<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0"><channel><title><![CDATA[johnettagaweraks]]></title><description><![CDATA[johnettagaweraks]]></description><link>https://johnettagaweraks.hashnode.dev</link><generator>RSS for Node</generator><lastBuildDate>Wed, 18 Nov 2026 15:54:58 GMT</lastBuildDate><atom:link href="https://johnettagaweraks.hashnode.dev/rss.xml" rel="self" type="application/rss+xml"/><language><![CDATA[en]]></language><ttl>60</ttl><item><title><![CDATA[Vitess VTAdmin: eleven title matches and an authentication layer you must design]]></title><description><![CDATA[VTAdmin is the web and API layer for managing a Vitess deployment. It is not a single monolithic console with a built-in identity system. The Vitess…]]></description><link>https://johnettagaweraks.hashnode.dev/vitess-vtadmin-eleven-title-matches-and-an-authentication-layer-you-must-design</link><guid isPermaLink="true">https://johnettagaweraks.hashnode.dev/vitess-vtadmin-eleven-title-matches-and-an-authentication-layer-you-must-design</guid><dc:creator><![CDATA[johnettagaweraks]]></dc:creator><pubDate>Wed, 18 Nov 2026 13:21:22 GMT</pubDate><content:encoded><![CDATA[<p>VTAdmin is the web and API layer for managing a Vitess deployment. It is not a single monolithic console with a built-in identity system. The Vitess documentation says authentication is obtained at the incoming HTTP and gRPC request boundary, authorization happens inside the <code>vtadmin.API</code> layer, and deployment-specific authentication must be designed by the operator. There is no trust boundary between <code>vtadmin-web</code> and <code>vtadmin-api</code>.</p>
<p>A count-only ZoomEye measurement on November 17, 2026 used <code>sub_type=all</code>, page 1, and a page size of 1. <code>app="Vitess"</code> returned 0. <code>title="VTAdmin"</code> returned 11. The title matches are a small candidate set for a management surface whose protection depends heavily on the surrounding architecture.</p>
<p>The phrase "no trust boundary" is not a vulnerability by itself. It is an architectural warning. The web client sends requests to the API, and the API must treat those requests according to the authentication and authorization policy. If the API is exposed directly, a proxy that protects the web page may not protect the management operations. If the API accepts requests from an internal network without authentication, an attacker who reaches that network can bypass the intended portal.</p>
<h2>What the counts can support</h2>
<p>Start an owned-asset review with the request path. Where does <code>vtadmin-web</code> run? Where does <code>vtadmin-api</code> listen? Which proxy or gateway sits in front of each? Is the API reachable from the same public host, a separate internal name, or a service mesh? Record the gRPC and HTTP endpoints separately. A title search may find only the web front end, while the API is the higher-value control plane.</p>
<p>Vitess provides role-based access control at the API layer. The documentation discusses authentication plugins and authorisation decisions based on the actor and resource. An operator must connect the actor to a trusted identity source and write rules that match the actual operational roles. An RBAC system with a permissive default or an actor derived from an unverified header is not a security boundary.</p>
<p>For each deployment, list the actors: human operators, CI jobs, monitoring systems, and support tools. Map each to the minimum API actions it needs. Review who can change keyspaces, shards, tablets, schema workflows, and topology records. Then verify that the API can distinguish those actors. A shared service account with broad authority makes incident attribution and least privilege difficult.</p>
<h2>Verification and next steps</h2>
<p>Network controls should reflect the control-plane role. Put VTAdmin behind an authenticated gateway or private management network. Restrict direct API access to expected proxy and operator networks. Use TLS for both browser and gRPC paths where the deployment supports it. Do not expose the API simply because the web client is served from the same hostname.</p>
<p>Passive evidence can include the page title, server header, TLS certificate, and response behaviour. It cannot prove that the API is open or that RBAC is weak. Do not send management actions to unknown instances. For an owned instance, use configuration and approved test accounts to verify that unauthenticated requests fail and that a low-privilege actor cannot perform administrative operations.</p>
<p>Logging should capture authentication results, the actor identity, the API resource and action, and the outcome. Correlate VTAdmin events with Vitess topology and schema changes. A missing actor field or a generic service identity makes an otherwise good authorisation decision hard to audit.</p>
<p>The 11 title matches are small enough for a focused review. The 0 application result may reflect incomplete fingerprint coverage or a naming difference. Neither result is the answer to whether the API is protected. The answer comes from the operator-designed authentication boundary and the actual RBAC rules.</p>
<p>VTAdmin is a useful control plane because it makes distributed Vitess operations visible. That same role makes it a high-value target. Inventory both web and API endpoints, attach them to a trusted identity source, and verify the path from browser to API. A management console without a clear trust boundary is an inventory gap waiting to become an incident.</p>
<h2>References</h2>
<ul>
<li><a href="https://vitess.io/docs/24.0/reference/vtadmin/role-based-access-control/">Vitess VTAdmin role-based access control</a></li>
<li><a href="https://vitess.io/docs/24.0/reference/vtadmin/operators_guide/">Vitess VTAdmin operator's guide</a></li>
</ul>
]]></content:encoded></item><item><title><![CDATA[Apache Polaris: four title matches and the identity boundary around a catalog API]]></title><description><![CDATA[Apache Polaris is a catalog service for lakehouse data. A reachable Polaris endpoint may expose namespaces, tables, principals, roles, grants, and the…]]></description><link>https://johnettagaweraks.hashnode.dev/apache-polaris-four-title-matches-and-the-identity-boundary-around-a-catalog-api</link><guid isPermaLink="true">https://johnettagaweraks.hashnode.dev/apache-polaris-four-title-matches-and-the-identity-boundary-around-a-catalog-api</guid><dc:creator><![CDATA[johnettagaweraks]]></dc:creator><pubDate>Wed, 18 Nov 2026 10:02:02 GMT</pubDate><content:encoded><![CDATA[<p>Apache Polaris is a catalog service for lakehouse data. A reachable Polaris endpoint may expose namespaces, tables, principals, roles, grants, and the management API used to change them. The security question is not whether a page says "Polaris" in its title. It is whether the catalog API is separated from the identities that can modify its security model.</p>
<p>A count-only ZoomEye request on November 18, 2026 returned four matches for <code>title="Apache Polaris"</code> and zero for <code>app="Apache Polaris"</code>. The request used page 1, page size 1, and <code>sub_type=all</code>. These are index observations, not four confirmed live catalogs and not four vulnerable systems. A title match can be a documentation page, a management console, or a stale response. A zero application-fingerprint count means the tested string did not classify an asset as that product.</p>
<p>The official Polaris documentation describes authentication through an external OpenID Connect identity provider as well as an internal authentication system. The API accepts bearer tokens. The management API is separate from the catalog data path and includes operations for principal roles and grants. Polaris also supports an external policy decision point such as Open Policy Agent.</p>
<h2>Identity is not authorization</h2>
<p>A valid token proves that an identity was issued by a trusted provider. It does not prove that the identity should be allowed to create a catalog, change a grant, or manage a principal. Polaris maps token claims to security roles and then applies catalog authorization. A deployment can therefore look correctly integrated with an identity provider while still granting too much authority to every authenticated user.</p>
<p>The first inventory step is to identify the token issuer and audience. Check that production tokens are issued for the production Polaris API, not accepted from a broad internal issuer without audience restrictions. Review the <code>root</code> principal and bootstrap credentials separately. The official guide describes using the bootstrap root token to create the initial principals and then avoiding root for ordinary API calls. If a deployment keeps using root for automation, the catalog has a single high-value identity with no useful separation.</p>
<p>The second step is to map principals to catalog objects. A catalog may contain several schemas and tables with different owners. A principal role that can create objects in one catalog should not automatically manage principals or grants in another. Export the role assignments and compare them with the intended data domains. Test a low-privilege user against each management endpoint rather than reading a login page.</p>
<p>The external policy decision point adds another boundary. A policy configuration that defaults to "none" for bearer authentication or uses an unreachable PDP can change how requests are evaluated. Policy changes should be versioned and tested in the same way as application code. The catalog should fail closed when an authorization decision cannot be made.</p>
<h2>What four title matches do and do not establish</h2>
<p>Four matches are a small observation set, so the safest use is as a probe for further validation. For each match, verify the service, certificate, response body, and path. Look for <code>/api/catalog</code> or management paths that answer without a 401 or 403. Do not send credentials to an unknown host and do not treat a login page as proof that the catalog API is protected.</p>
<p>Use the count to ask an architecture question: do we know every catalog deployment in our estate? The public count is only a hint about how often the product name appears in indexed web content. Internal inventory, cloud resource lists, and Kubernetes discovery are more reliable for owned assets.</p>
<p>The defensible conclusion is narrow. Polaris separates authentication from authorization and supports external identity and policy components, which gives operators useful controls. The risky condition is a deployment that treats a valid token as blanket permission or exposes a management API without a strong network boundary. The four title matches warrant a controlled inventory check, not a claim of compromise.</p>
<h2>Query evidence</h2>
<pre><code class="language-text">title="Apache Polaris"
sub_type=all, page=1, pagesize=1, observed 2026-11-18T07:42:21+08:00, count=4
</code></pre>
<h2>References</h2>
<ul>
<li><a href="https://polaris.apache.org/in-dev/unreleased/managing-security/">Apache Polaris: Managing Security</a></li>
<li><a href="https://polaris.apache.org/in-dev/unreleased/polaris-api-specs/">Apache Polaris: API Reference</a></li>
<li><a href="https://polaris.apache.org/in-dev/unreleased/managing-security/external-pdp/">Apache Polaris: External Policy Decision Point</a></li>
<li><a href="https://www.zoomeye.ai/searchResult?q=dGl0bGU9IkFwYWNoZSBQb2xhcmlzIg%3D%3D">ZoomEye count-only search for Apache Polaris</a></li>
</ul>
]]></content:encoded></item><item><title><![CDATA[CAEP Turns Identity Changes into Access Events, but Consumers Must Enforce Them]]></title><description><![CDATA[Continuous Access Evaluation Profile 1.0 defines how cooperating providers send security events about a subject. The events can describe a session…]]></description><link>https://johnettagaweraks.hashnode.dev/caep-turns-identity-changes-into-access-events-but-consumers-must-enforce-them</link><guid isPermaLink="true">https://johnettagaweraks.hashnode.dev/caep-turns-identity-changes-into-access-events-but-consumers-must-enforce-them</guid><dc:creator><![CDATA[johnettagaweraks]]></dc:creator><pubDate>Wed, 18 Nov 2026 07:25:52 GMT</pubDate><content:encoded><![CDATA[<p>Continuous Access Evaluation Profile 1.0 defines how cooperating providers send security events about a subject. The events can describe a session revocation, a credential change, a token revocation, an assurance-level change or another condition that should alter access. The profile is built on the Shared Signals Framework, so it combines event semantics with a delivery and subscription model.</p>
<p>The first engineering mistake is to treat event delivery as authorization. A receiver that accepts an event without checking the issuer, signature, subject identifier and event type has created a new unauthenticated API. The second mistake is to accept the event and never apply it to a live session. A real CAEP deployment needs both a trustworthy event channel and a local policy that knows which resources must react.</p>
<h2>The event is not the decision</h2>
<p>A transmitter may report that a user's session was revoked. The application should decide whether that revocation applies to its own session, its own token cache and its own high-risk operations. Session identifiers are deployment-specific. A subject identifier may be an email address in one provider and an opaque pairwise identifier in another. The receiver needs a mapping that is stable, auditable and resistant to account reuse.</p>
<p>The event payload also has a timestamp, issuer and event type. These fields should be checked before the event reaches the authorization cache. A delayed event can be useful for investigation but may be too old to override a more recent authentication. A duplicate event should be idempotent. An event for an unknown subject should be quarantined, not silently applied to a similarly named user.</p>
<p>CAEP is most valuable when the access decision is close to the resource. An API gateway can reject a new request after a revocation event. A long-lived background job may already be executing with a token that was valid when the job started. The design should define whether that job is interrupted, allowed to finish with an audit record, or required to reauthorize before its next high-risk action.</p>
<h2>Build the trust path before the dashboard</h2>
<p>The receiver should have a configured list of trusted transmitters, a key set, an accepted event schema and a maximum clock skew. It should verify signatures with a library that rejects algorithm changes and unknown key identifiers. It should authenticate the event source separately from the subject event. A signature that proves the payload came from a transmitter does not prove that the transmitter is authorized to make that claim.</p>
<p>Subscriptions should be reviewed like any other privileged integration. Record which events are requested, which tenant or organization identifiers are in scope, who owns the subscription and how keys rotate. A test event can show that the pipe works, but it does not show that the receiver's policy is correct.</p>
<p>Log the event identifier, issuer, type, subject mapping, receive time, verification result and action taken. Protect the subject identifiers and avoid copying raw identity claims into general-purpose analytics. Keep a separate audit trail for policy overrides, because the most important event is often the one an operator chose not to apply.</p>
<h2>Test the worst moments</h2>
<p>Run failure tests during a credential reset, a device loss and a session invalidation. Confirm that a revoked session cannot start a new privileged operation, that a cached access token is invalidated within the stated window, and that an event from an unknown issuer is rejected. Then test a delayed or duplicated event. The system should not revoke an unrelated user or restore an authority that a later event removed.</p>
<p>The profile gives identity providers and applications a common vocabulary for continuous evaluation. The organization still has to decide which decisions are continuous, which events can override which sessions and how quickly a local cache must converge. Those decisions belong in the access policy, not in the event feed.</p>
<h2>References</h2>
<ul>
<li><a href="https://openid.net/specs/openid-caep-1_0-final.html">OpenID Continuous Access Evaluation Profile 1.0</a></li>
<li><a href="https://openid.net/specs/openid-sharedsignals-framework-1_0-final.html">OpenID Shared Signals Framework 1.0</a></li>
</ul>
]]></content:encoded></item><item><title><![CDATA[OpenVINO Model Server Has No Application Fingerprint and Twenty Broad Title Matches]]></title><description><![CDATA[OpenVINO Model Server serves machine-learning models through KServe-compatible gRPC and REST endpoints. It can run as a container or as a service, and its…]]></description><link>https://johnettagaweraks.hashnode.dev/openvino-model-server-has-no-application-fingerprint-and-twenty-broad-title-matches</link><guid isPermaLink="true">https://johnettagaweraks.hashnode.dev/openvino-model-server-has-no-application-fingerprint-and-twenty-broad-title-matches</guid><dc:creator><![CDATA[johnettagaweraks]]></dc:creator><pubDate>Wed, 18 Nov 2026 05:21:13 GMT</pubDate><content:encoded><![CDATA[<p>OpenVINO Model Server serves machine-learning models through KServe-compatible gRPC and REST endpoints. It can run as a container or as a service, and its security model depends heavily on the deployment around it. The documentation states that the default container runs as a local account with UID 5000. It also states that the server does not currently provide access restrictions or traffic encryption on its gRPC and REST API endpoints.</p>
<p>ZoomEye count-only queries on 2026-11-18 returned zero matches for <code>app="OpenVINO Model Server"</code> and zero for <code>title="OpenVINO Model Server"</code>. A shorter application query, <code>app="OpenVINO"</code>, also returned zero. The broad title query <code>title="OpenVINO"</code> returned 20. Those 20 matches are not a model-server census. They may include project pages, demos, documentation or applications that use the OpenVINO name without exposing the server API.</p>
<p>The result is a useful example of fingerprint absence. A product can be present while ZoomEye has no stable application signature for the deployment. The absence of <code>app</code> data should not become a statement that no instances are exposed. It means the query cannot support that conclusion.</p>
<h2>The deployment must supply the missing control</h2>
<p>The documentation's security guidance is direct: restrict network access and use a reverse proxy or gateway when authentication or encryption is required. A deployment that publishes the REST or gRPC port to the internet without that layer has no product-level authentication to fall back on. The gateway should terminate TLS, authenticate clients and restrict which model endpoints are visible.</p>
<p>Model content adds another concern. The server may load model files from a local directory or a remote repository. A model can contain custom code, pre- and post-processing logic or configuration that changes how input is handled. Treat model storage as a trusted supply-chain input. Do not let an unauthenticated client select an arbitrary repository path or model version.</p>
<p>Container hardening should be part of the deployment template. The documented non-root UID is a useful default, but it does not replace a read-only model mount, a restrictive network policy or a resource limit. If a model needs a writable directory, give it a dedicated volume with a defined lifecycle. Do not run the container with additional capabilities unless the model or runtime produces a specific, tested requirement.</p>
<h2>Look for the gateway, not the brand</h2>
<p>The practical inventory question is not only "where is OpenVINO?" It is "which HTTP and gRPC services are running inference and what protects them?" Search service inventories for OpenVINO base images, container names and model repository paths. Correlate those findings with Kubernetes ingress, service mesh policy and external DNS.</p>
<p>For each deployment, record the model source, the network path, the TLS terminator, the authentication method and the identity used to call the model. Test an unauthenticated request from a network that should be denied. Test a request with a valid identity for a model the identity should not access. A successful inference call is the useful negative result here.</p>
<p>The 20 broad title matches do not establish exposure. The official security guidance does establish the consequence of leaving the API unprotected. OpenVINO Model Server is best treated as an internal inference component with a gateway in front, not as a self-authenticating web application.</p>
<h2>References</h2>
<ul>
<li><a href="https://docs.openvino.ai/nightly/model-server/ovms_docs_security.html">OpenVINO Model Server security considerations</a></li>
<li><a href="https://docs.openvino.ai/2025/model-server/ovms_what_is_openvino_model_server.html">OpenVINO Model Server overview</a></li>
<li><a href="https://docs.openvino.ai/2024/openvino-workflow/model-server/ovms_extras_nginx-mtls-auth-readme.html">Securing OpenVINO Model Server with NGINX</a></li>
</ul>
]]></content:encoded></item><item><title><![CDATA[Onyx AI at 102 title matches: API keys connect users, agents, and enterprise data]]></title><description><![CDATA[Onyx is an open-source enterprise search and AI assistant that can connect to internal documents, applications, and chat workflows. A self-hosted instance…]]></description><link>https://johnettagaweraks.hashnode.dev/onyx-ai-at-102-title-matches-api-keys-connect-users-agents-and-enterprise-data</link><guid isPermaLink="true">https://johnettagaweraks.hashnode.dev/onyx-ai-at-102-title-matches-api-keys-connect-users-agents-and-enterprise-data</guid><dc:creator><![CDATA[johnettagaweraks]]></dc:creator><pubDate>Wed, 18 Nov 2026 00:00:35 GMT</pubDate><content:encoded><![CDATA[<p>Onyx is an open-source enterprise search and AI assistant that can connect to internal documents, applications, and chat workflows. A self-hosted instance is more than a search page. It can hold connectors, user groups, model-provider credentials, API keys, agent definitions, and actions that call other services. The external exposure question should follow those identities.</p>
<p>ZoomEye returned 102 matches for <code>title="Onyx AI"</code> and zero for <code>app="Onyx AI"</code> on November 18, 2026, using <code>sub_type=all</code>, page 1, and page size 1. The count is an index observation. It may include product pages and unrelated content, and it does not show whether any host is a live or misconfigured deployment.</p>
<p>Onyx's official developer page says the platform supports fine-grained API keys and APIs for messages, agents, LLMs, actions, and user authentication. Its product material describes permission-aware retrieval across connectors and support for self-hosted, cloud, or air-gapped deployment. Those features make identity integration central: the assistant has to decide which user can see which source, and which agent or API key can call which action.</p>
<h2>Three identities in one deployment</h2>
<p>The first identity is the human user. Group membership and source permissions determine which documents the assistant can retrieve. A public chat link or an unauthenticated search endpoint can bypass that model if it exposes results without the same authorization check.</p>
<p>The second identity is the API key. A key may be scoped to a user, a service, or an integration. The important question is whether it inherits the creator's access or has a narrower purpose. An API key that can create agents, add connectors, or inspect messages has more authority than a read-only chat token. Inventory keys by owner, scope, creation date, and last use.</p>
<p>The third identity is the agent or action. A connector may hold OAuth tokens for Drive, Slack, email, or another repository. An action may have access to a ticketing system or internal API. When a user asks the assistant to perform a task, the action should know whether it is acting for that user, as a shared service, or as an administrator. The presence of an LLM in the request path does not remove the need for object-level authorization.</p>
<h2>What to inspect on an owned instance</h2>
<p>Start from the application configuration and database, not from a title search. List users, groups, connectors, service accounts, API keys, agents, actions, and model providers. Check whether the deployment allows public signup or guest access. Review OAuth scopes and revoke connectors that no longer have an owner. Confirm that a low-privilege user cannot enumerate documents outside the connected source permissions.</p>
<p>Test the API separately from the browser. A request with no token should receive an authentication error, not a generic page that reveals deployment details. A token for one user should not read another user's chat. A connector credential should not be reusable as a general API token. If the application supports administrative actions, require a distinct role and log the change.</p>
<p>The language model provider adds another secret. A shared model key may be billed to the organization and may allow access to provider features that are not intended for every user. Use per-environment keys, spending limits, and rotation. Do not place the provider key in a prompt, a document, or a connector configuration that ordinary users can read.</p>
<h2>Turning a broad count into a controlled check</h2>
<p>The exact title phrase is useful because it is specific, but it still cannot distinguish a legitimate public product page from an internal assistant. Before requesting any response body, verify that the host belongs to the organization. Respect scope and do not attempt authentication against systems you do not own. For owned hosts, compare the public endpoint with the internal inventory and the expected deployment mode.</p>
<p>The result of the check should be a list of identities. Which human users can reach the assistant? Which keys can change its connectors? Which agents can call external tools? Which data sources trust the assistant as a user? A deployment that answers these questions has a meaningful security boundary around the product name.</p>
<p>One hundred two title matches are small enough to review manually but large enough to show why automated classification can mislead. The product's value comes from joining systems together. Its risk comes from allowing one key or one agent to join them too broadly.</p>
<h2>Query evidence</h2>
<pre><code class="language-text">title="Onyx AI"
sub_type=all, page=1, pagesize=1, observed 2026-11-18T07:42:27+08:00, count=102
</code></pre>
<h2>References</h2>
<ul>
<li><a href="https://onyx.app/developer-platform">Onyx developer platform</a></li>
<li><a href="https://onyx.app/">Onyx product overview</a></li>
<li><a href="https://www.zoomeye.ai/searchResult?q=dGl0bGU9Ik9ueXggQUki">ZoomEye count-only search for Onyx AI</a></li>
</ul>
]]></content:encoded></item><item><title><![CDATA[The leaf CRL check that never runs in GoAhead]]></title><description><![CDATA[CVE-2026-115881 is a precise certificate-state failure in GoAhead's OpenSSL implementation. The server loads a CRL but does not set the OpenSSL…]]></description><link>https://johnettagaweraks.hashnode.dev/the-leaf-crl-check-that-never-runs-in-goahead</link><guid isPermaLink="true">https://johnettagaweraks.hashnode.dev/the-leaf-crl-check-that-never-runs-in-goahead</guid><dc:creator><![CDATA[johnettagaweraks]]></dc:creator><pubDate>Tue, 17 Nov 2026 20:42:07 GMT</pubDate><content:encoded><![CDATA[<h2>Vulnerability overview</h2>
<p>CVE-2026-115881 is a precise certificate-state failure in GoAhead's OpenSSL implementation. The server loads a CRL but does not set the OpenSSL verification flag that makes the list effective. An affected mutual TLS service accepts a revoked client certificate. EmbedThis assigns CVSS 3.1 score 9.1 and fixes the issue in GoAhead 6.1.1.</p>
<p>The affected versions are 3.4.5 through 6.1.0. The vendor maps the issue to CWE-295 and CWE-299. The first describes improper certificate validation; the second describes an improper check for certificate revocation.</p>
<h2>Mechanism and exploitation conditions</h2>
<p>The relevant setting is <code>ME_GOAHEAD_SSL_REVOKE</code>, which points to a PEM-encoded CRL. When the path is non-empty, GoAhead obtains the OpenSSL certificate store and calls <code>X509_STORE_load_locations</code>. A failed load aborts startup, so the file-loading operation is not the missing step.</p>
<p>The missing step is a call to enable <code>X509_V_FLAG_CRL_CHECK</code>. OpenSSL's verifier checks the flag before it evaluates revocation. If neither CRL checking nor OCSP checking is enabled, the revocation routine returns success without consulting the list. The certificate can be listed as revoked and still pass.</p>
<p>GoAhead's verification callback contains a default path that would reject <code>X509_V_ERR_CERT_REVOKED</code>, but the verifier never produces that error. A valid chain with a good signature and a revoked leaf can therefore look successful to the callback.</p>
<p>Exploitation requires the OpenSSL backend, <code>ME_GOAHEAD_SSL_VERIFY_PEER=1</code>, a configured CA bundle, <code>ME_GOAHEAD_SSL_VERIFY_ISSUER=1</code>, a non-empty CRL, and a revoked certificate from a trusted CA. The default build does not enable the first steps.</p>
<h2>Impact</h2>
<p>The attacker receives the access attached to the certificate identity. If the certificate represented a field device, a support account, or an administrator endpoint, the application may expose management routes or operations that the operator believed had been revoked.</p>
<p>The CVSS vector is network reachable, low complexity, no privileges, no user interaction, scope unchanged, with high confidentiality and integrity impact and no availability impact. It assumes the attacker already has the revoked certificate. The severity comes from the failure to enforce an expected access boundary.</p>
<h2>Affected products and scope</h2>
<p>The defect is in GoAhead's OpenSSL integration. Mbedtls builds use a different revocation path and the vendor says they are not vulnerable to this CVE. GoAhead appears in many embedded management systems, so the product build and certificate configuration are more useful than a bare product name.</p>
<p>The affected range is 3.4.5 through 6.1.0. Patching decisions should include the firmware version, the TLS backend, whether mutual TLS is required, the trusted CA bundle, issuer verification settings, and how the CRL is delivered and refreshed.</p>
<h2>Exposure context</h2>
<p>On November 18, 2026, ZoomEye returned 6,627,002 product-fingerprint matches for <code>app="GoAhead"</code> and 0 matches for the exact CVE query. Exposure uses the larger successful product count, but that count does not show backend or certificate settings.</p>
<p>The <a href="https://www.zoomeye.ai/searchResult?q=YXBwPSJHb0FoZWFkIg%3D%3D">ZoomEye result</a> should be treated as an inventory signal. It cannot establish that a device has a trusted CA bundle or a revoked client credential.</p>
<h2>Remediation and mitigations</h2>
<p>Install GoAhead 6.1.1 or later. The fixed code enables leaf-certificate CRL checking after the list loads. The vendor deliberately leaves chain-wide checking disabled because it requires CRLs for every issuer in the trust chain.</p>
<p>Until the fixed firmware is available, restrict access to the management service and review whether mutual TLS can be disabled safely. A network control can reduce reachability, but it does not repair the server's certificate decision.</p>
<p>Test the patch with a certificate that the managed CA has revoked. Check for a failed handshake and the expected verification error. Also test a valid certificate to ensure the update did not break normal client authentication. The meaningful result is that the loaded CRL now participates in verification.</p>
<h2>References</h2>
<ul>
<li><a href="https://github.com/embedthis/goahead/security/advisories/GHSA-5m75-xq69-c3c5">EmbedThis advisory GHSA-5m75-xq69-c3c5</a></li>
<li><a href="https://github.com/embedthis/goahead/releases/tag/v6.1.1">GoAhead v6.1.1 release notes</a></li>
<li><a href="https://foxio.io/blog/choose-your-backend-three-tls-bugs-in-goahead">FoxIO technical research</a></li>
<li><a href="https://www.openwall.com/lists/oss-security/2026/10/28/1">oss-security disclosure</a></li>
</ul>
]]></content:encoded></item><item><title><![CDATA[From scanner traffic to an interactive YOOtheme operator]]></title><description><![CDATA[The Unit 42 record for CVE-2026-102785 shows two stages that defenders often handle as separate events: automated exploitation and later hands-on activity…]]></description><link>https://johnettagaweraks.hashnode.dev/from-scanner-traffic-to-an-interactive-yootheme-operator</link><guid isPermaLink="true">https://johnettagaweraks.hashnode.dev/from-scanner-traffic-to-an-interactive-yootheme-operator</guid><dc:creator><![CDATA[johnettagaweraks]]></dc:creator><pubDate>Tue, 17 Nov 2026 17:21:27 GMT</pubDate><content:encoded><![CDATA[<p>The Unit 42 record for CVE-2026-102785 shows two stages that defenders often handle as separate events: automated exploitation and later hands-on activity. The sequence began with a scanner writing JPEG/PHP polyglot webshells to a WordPress uploads directory. Each drop was checked within seconds. About two minutes later, an interactive operator uploaded a fuller ALFA Team webshell and attempted to install gsocket.</p>
<p>The vulnerability affects YOOtheme Pro 5.0.0 through 5.0.52 and is fixed in 5.0.53. Its image handler accepts a public signed image URL and insufficiently binds additional operations and the output destination to the signature. The attacker can therefore control file content and placement without authenticating to WordPress or Joomla.</p>
<h2>Why the two-minute gap matters</h2>
<p>Fast validation indicates that the first stage was built for scale. A scanner can send many requests, check the result, and move on. The later operator step indicates that at least one successful drop was noticed and developed into broader access. The transition can happen faster than a daily vulnerability-management cycle.</p>
<p>Logs need to preserve enough context to connect the stages. A scanner request may return HTTP 500 while still writing the file. The first-stage shell can then respond with a marker such as <code>WPCONFIG</code> and accept commands through a query parameter. The operator may use different source addresses, so an IP-only pivot can split what should be one investigation.</p>
<h2>Evidence to correlate</h2>
<p>Start with the request URI. Look for <code>yootheme=cache/</code>, <code>FromFile=php://input</code>, and <code>Save=</code> values ending in <code>.php</code>. Include 500 responses. Correlate those times with new <code>yt[0-9a-f]{7}.php</code> files and with execution by the web-server process.</p>
<p>On Windows, process telemetry may show the web server spawning <code>cmd.exe /s /c</code> for an <code>echo YTS...</code> marker followed by <code>id</code>. The operator later ran generic Linux commands despite the target being a Windows host, which Unit 42 interpreted as tooling aimed broadly at WordPress sites. The ALFA kit appeared under <code>wp-content/uploads/ALFA_DATA/alfacgiapi/</code>.</p>
<p>Network telemetry may show the shell downloading a remote PHP loader or using <code>curl</code> to fetch and run gsocket. Those events occur after the initial file write. If an endpoint solution blocks one step, the other steps can still provide evidence that the web shell executed.</p>
<h2>Containment without losing the trail</h2>
<p>When the request and execution evidence align, preserve the web log, process record, file, and payload hash. Isolate the host. Do not overwrite the uploads directory before capture. The affected application may be only the entry point, so review database access, administrator sessions, deployment secrets, and outbound connections from the same window.</p>
<p>Updating YOOtheme Pro to 5.0.53 or later prevents new writes through this path. It does not invalidate a shell already on disk. If the server executed an attacker-controlled file, rebuild or restore from a clean backup and rotate every credential the web process could read.</p>
<h2>Exposure context</h2>
<p>ZoomEye's <code>http.body="YOOtheme"</code> search returned 34,682 pages, while the exact CVE filter returned zero. The product-string result is a broad inventory lead and not a count of vulnerable or compromised systems. It should be combined with version checks and endpoint evidence.</p>
<p>The broader lesson is that exploitation timelines now overlap automated scanning and human follow-up. Detection for CVE-2026-102785 should not stop at a blocked request. It should look for the successful side effect, the validation command, and the later operator activity that turns a file write into a durable intrusion.</p>
<h2>References</h2>
<ul>
<li><a href="https://www.cve.org/CVERecord?id=CVE-2026-102785">CVE record for CVE-2026-102785</a></li>
<li><a href="https://nvd.nist.gov/vuln/detail/CVE-2026-102785">NVD entry for CVE-2026-102785</a></li>
<li><a href="https://yootheme.com/changelog">YOOtheme Pro changelog</a></li>
<li><a href="https://github.com/PaloAltoNetworks/Unit42-timely-threat-intel/blob/main/2026-10-26-exploitation-of-CVE-2026-102785.txt">Unit 42 timely threat intelligence</a></li>
</ul>
]]></content:encoded></item><item><title><![CDATA[Jahia returned eight titles and no application matches, so the JCR and token paths lead]]></title><description><![CDATA[Jahia is a Java content platform with site administration, server administration, a content repository, REST access, and optional personal API tokens. A…]]></description><link>https://johnettagaweraks.hashnode.dev/jahia-returned-eight-titles-and-no-application-matches-so-the-jcr-and-token-paths-lead</link><guid isPermaLink="true">https://johnettagaweraks.hashnode.dev/jahia-returned-eight-titles-and-no-application-matches-so-the-jcr-and-token-paths-lead</guid><dc:creator><![CDATA[johnettagaweraks]]></dc:creator><pubDate>Tue, 17 Nov 2026 14:02:00 GMT</pubDate><content:encoded><![CDATA[<p>Jahia is a Java content platform with site administration, server administration, a content repository, REST access, and optional personal API tokens. A small public title result can still identify a high-value deployment. The review should move from the page name to the repository and token paths that can change content or administer the server.</p>
<p>On 2026-11-17, a bounded ZoomEye SDK query used <code>sub_type=all</code>, page 1, and <code>pagesize=1</code>. <code>app="Jahia"</code> returned 0 total matches at 2026-11-17T06:55:11.495164+08:00. <code>title="Jahia"</code> returned 8 total matches at 2026-11-17T06:55:12.681309+08:00. The observed total is shown by the <a href="https://www.zoomeye.ai/searchResult?q=YXBwPSJKYWhpYSI%3D">application query</a> and the <a href="https://www.zoomeye.ai/searchResult?q=dGl0bGU9IkphaGlhIg%3D%3D">title query</a>. The page size limited returned records, not the matched total.</p>
<p>The app query returned zero and the title query returned eight. The small sample is manageable, but it is not an exhaustive count of Jahia deployments. A public host may use a custom title, a gateway, a headless front end, or an older registration. The zero application result should be stored with the title result and labelled as a field-specific observation.</p>
<h2>Separate site and server administration</h2>
<p>Jahia documentation separates site administration from server administration and describes roles that grant access to the backoffice and server configuration. A site editor should not inherit server-level module or user administration. The review should identify who can change roles, install modules, use the JCR API, and create personal API tokens.</p>
<p>Personal API tokens are credentials that may work across a broad set of endpoints. The Jahia personal API tokens documentation describes restricted token-authentication URL patterns and permissions for token administration. A deployment should restrict token usage to the endpoints an integration needs, name the owner, and remove tokens when the owner leaves. A token created for a one-time migration should not remain as a permanent backoffice credential.</p>
<h2>Check the content repository boundary</h2>
<p>The JCR API can reach content that the public site does not render. The same authentication session may work in a browser and in a REST call, which makes proxy and session policy important. A gateway that protects the backoffice HTML but leaves the JCR endpoint exposed has protected one representation of the same authority. Verify the route coverage, HTTP methods, authorization checks, and response handling for all content APIs.</p>
<p>Also review module permissions. A module can add a workflow, a connector, or a custom endpoint that was not part of the core permission test. A public deployment with few visible pages can still have many internal sites or custom modules. The owner should provide that inventory.</p>
<h2>Turn the title list into a bounded review</h2>
<p>For each title match, record host, title, certificate, owner, and expected deployment. Ask whether the host is a site, an authoring instance, a preview environment, or a shared platform. Do not test user credentials or token creation during external discovery. If the host is confirmed, review its role assignments and token list through an authorised administrative process.</p>
<p>The query result can be reproduced by storing the exact dork, scope, count, and time. A later change from eight titles to a different number should be explained before it changes the asset ledger. A title change can be a rebrand, a login-page update, or a new deployment. It does not reveal which one without owner evidence.</p>
<h2>Keep tokens out of the measurement</h2>
<p>Never paste a personal API token into a ZoomEye query or a public report. The index count does not need the secret. If a token is found in a public page or repository, treat it as a disclosure: identify the owning service, revoke it through the product, check its permission and usage, and preserve evidence for the incident process.</p>
<p>Jahia's eight-title result is not a large population, but the assets may hold the site's content and administration path. Small counts make ownership review practical; they do not make the authority behind the host small.</p>
<h2>References</h2>
<p>[1] ZoomEye application query. <a href="https://www.zoomeye.ai/searchResult?q=YXBwPSJKYWhpYSI%3D">https://www.zoomeye.ai/searchResult?q=YXBwPSJKYWhpYSI%3D</a></p>
<p>[2] ZoomEye title query. <a href="https://www.zoomeye.ai/searchResult?q=dGl0bGU9IkphaGlhIg%3D%3D">https://www.zoomeye.ai/searchResult?q=dGl0bGU9IkphaGlhIg%3D%3D</a></p>
<p>[3] Jahia server administration. <a href="https://academy.jahia.com/documentation/jahia-cms/jahia-8.2/system-administrator/administering-your-server/about-server-administration">https://academy.jahia.com/documentation/jahia-cms/jahia-8.2/system-administrator/administering-your-server/about-server-administration</a></p>
<p>[4] Jahia personal API token usage. <a href="https://github.com/Jahia/personal-api-tokens/blob/main/docs/USAGE.md">https://github.com/Jahia/personal-api-tokens/blob/main/docs/USAGE.md</a></p>
]]></content:encoded></item><item><title><![CDATA[Shared compression dictionaries add cache and secret-dependent state]]></title><description><![CDATA[Compression Dictionary Transport lets a server compress an HTTP response using a dictionary downloaded earlier. The browser can advertise a matching…]]></description><link>https://johnettagaweraks.hashnode.dev/shared-compression-dictionaries-add-cache-and-secret-dependent-state</link><guid isPermaLink="true">https://johnettagaweraks.hashnode.dev/shared-compression-dictionaries-add-cache-and-secret-dependent-state</guid><dc:creator><![CDATA[johnettagaweraks]]></dc:creator><pubDate>Tue, 17 Nov 2026 10:41:34 GMT</pubDate><content:encoded><![CDATA[<p>Compression Dictionary Transport lets a server compress an HTTP response using a dictionary downloaded earlier. The browser can advertise a matching dictionary, the server can refer to a dictionary identifier, and both sides save bandwidth when a site serves many similar resources. The feature also adds a long-lived state object that affects encoding, caching, and information leakage.</p>
<p>RFC 9842 defines the transport and its security considerations. The Risks are not limited to a broken decompressor. Shared dictionaries can make compression side channels easier to vary or repeat. If a response mixes attacker-controlled input with a secret, the compressed size can reveal information about the secret even when the dictionary itself contains only public data. The RFC recommends preventing active use of multiple dictionaries for the same content, disabling compression when any part of the content comes from an uncontrolled source, and protecting dictionary access with the same care as the response.</p>
<h2>Same-origin restrictions are necessary</h2>
<p>Browser implementations restrict dictionaries to the same origin and tie dictionary-compressed resources to the document origin or a valid CORS path. The dictionary is also subject to cache partitioning. These restrictions reduce cross-origin reuse and tracking, but they do not remove the need for a server policy. A compromised same-origin resource can still influence what a user receives.</p>
<p>The dictionary's URL, identifier, and matching rules should be treated as configuration, not optional performance metadata. A deployment that accepts a dictionary identifier from an untrusted request without checking its scope may let a client steer the server toward a stale or attacker-influenced object. The server should validate the exact identifier and match rule, and it should fail closed when the expected dictionary is unavailable.</p>
<h2>Plan versioning and rollback</h2>
<p>Use content hashes or monotonic versions for dictionary identity. Keep an old dictionary available until active clients have refreshed, then remove it through a controlled cache policy. Test a dictionary miss, a stale identifier, a corrupt payload, and a response where the encoded size differs from the stored metadata. A fast rollback should restore a known good dictionary without changing the response semantics.</p>
<p>Because the feature is controlled by HTTP headers and sometimes HTML relationship attributes, review Content Security Policy and proxy rewriting rules. A dictionary request that is blocked by <code>connect-src</code> or a gateway can produce a performance regression that looks like a network fault. Log the dictionary identifier and result, not the sensitive response content.</p>
<h2>Do not compress mixed trust</h2>
<p>The most important rule is simple: do not combine secrets from one user with input chosen by another user in a dictionary-compressed response. If a template must mix public and private data, separate the response or disable dictionary compression for that route. Measure the benefit against the new state and side-channel surface. Compression is a transport optimization; it is not a reason to weaken a response boundary.</p>
<h2>References</h2>
<p>[1] RFC 9842, Compression Dictionary Transport. <a href="https://datatracker.ietf.org/doc/rfc9842/">https://datatracker.ietf.org/doc/rfc9842/</a></p>
<p>[2] MDN, Compression Dictionary Transport. <a href="https://developer.mozilla.org/en-US/docs/Web/HTTP/Guides/Compression_dictionary_transport">https://developer.mozilla.org/en-US/docs/Web/HTTP/Guides/Compression_dictionary_transport</a></p>
<p>[3] MDN, Use-As-Dictionary. <a href="https://developer.mozilla.org/en-US/docs/Web/HTTP/Reference/Headers/Use-As-Dictionary">https://developer.mozilla.org/en-US/docs/Web/HTTP/Reference/Headers/Use-As-Dictionary</a></p>
]]></content:encoded></item><item><title><![CDATA[An SSH agent can hold a key without granting every destination]]></title><description><![CDATA[An SSH agent is convenient because a private key can stay in one process while many terminals request signatures. The usual risk is broad: if every client…]]></description><link>https://johnettagaweraks.hashnode.dev/an-ssh-agent-can-hold-a-key-without-granting-every-destination</link><guid isPermaLink="true">https://johnettagaweraks.hashnode.dev/an-ssh-agent-can-hold-a-key-without-granting-every-destination</guid><dc:creator><![CDATA[johnettagaweraks]]></dc:creator><pubDate>Tue, 17 Nov 2026 07:03:17 GMT</pubDate><content:encoded><![CDATA[<p>An SSH agent is convenient because a private key can stay in one process while many terminals request signatures. The usual risk is broad: if every client can ask the agent for the same key, a compromised jump host or session inherits access to every server that accepts it. OpenSSH added destination-constrained keys to narrow that delegation, but the constraint only helps when the key is loaded and managed correctly.</p>
<p><a href="https://man.openbsd.org/ssh-add.1">ssh-add(1)</a> documents the <code>-h destination_constraint</code> option. The OpenSSH project's <a href="https://www.openbsd.org/openssh/agent-restrict.html">agent-restriction design</a> says destination constraints were added in OpenSSH 8.9 and that keys are not constrained by default. When a constrained key is added, <code>ssh-add</code> looks up the destination host keys in the local <code>known_hosts</code> database and sends those restrictions to the agent.</p>
<h2>How the restriction works</h2>
<p>A destination constraint tells the agent that a key is intended for a particular host. The agent checks the signing request against the host-key identity it associated with the key. This is useful when an operator needs to use a key to a bastion or a named set of servers without making the same key usable for every connection made through the agent socket.</p>
<p>The first operational condition is a trustworthy host-key database. An unknown host cannot be safely encoded as a destination constraint. The agent also needs a current entry when a host key is rotated. Teams should provision and update <code>known_hosts</code> through configuration management instead of approving an unexpected key during an incident.</p>
<p>A constrained key does not remove the need for server-side authorization. The destination must still validate the authenticated user, enforce the <code>authorized_keys</code> options, and record the session. The constraint also does not make a stolen forwarding socket harmless if the same agent holds unconstrained keys. Inventory every loaded key, not only the one being tested.</p>
<h2>Deployment pattern</h2>
<p>Use separate keys for separate trust domains. Load the narrow key with one or more explicit <code>-h</code> values, keep its lifetime short, and avoid <code>ssh-add</code> defaults that turn a temporary key into a permanent one. For automation, prefer a dedicated agent socket or an SSH configuration that restricts the identity to the intended host. Remove the key when the maintenance task ends.</p>
<p>Review client configuration for <code>ForwardAgent</code>. Forwarding exposes the agent socket to the remote host, and destination constraints are the control that determines what a signing request from that side can reach. Log the host-key fingerprint used for the constraint and the identity of the person or job that loaded it.</p>
<p>A practical test should connect to an allowed destination and to a second destination that the key must not reach. Check the agent's key listing, the latest <code>known_hosts</code> entry, and the SSH server's authentication log. A failed connection with a clear agent denial is a better result than a successful login that reveals the constraint was never loaded.</p>
<p>OpenSSH destination constraints make delegated signing authority narrower. They are a useful second control behind host-key verification, least-privilege server accounts, and prompt removal of temporary credentials.</p>
<h2>References</h2>
<ul>
<li><a href="https://man.openbsd.org/ssh-add.1">OpenBSD manual: ssh-add(1)</a></li>
<li><a href="https://www.openbsd.org/openssh/agent-restrict.html">OpenSSH: SSH agent restriction</a></li>
</ul>
]]></content:encoded></item><item><title><![CDATA[One NanoMQ title match still points to an MQTT and HTTP control plane]]></title><description><![CDATA[A small internet count can describe a narrow fingerprint rather than a harmless service. NanoMQ is a lightweight MQTT broker with authentication…]]></description><link>https://johnettagaweraks.hashnode.dev/one-nanomq-title-match-still-points-to-an-mqtt-and-http-control-plane</link><guid isPermaLink="true">https://johnettagaweraks.hashnode.dev/one-nanomq-title-match-still-points-to-an-mqtt-and-http-control-plane</guid><dc:creator><![CDATA[johnettagaweraks]]></dc:creator><pubDate>Tue, 17 Nov 2026 03:21:21 GMT</pubDate><content:encoded><![CDATA[<p>A small internet count can describe a narrow fingerprint rather than a harmless service. NanoMQ is a lightweight MQTT broker with authentication, access-control, and HTTP API options. A count-only ZoomEye probe found one asset with <code>title="NanoMQ"</code> and zero with <code>app="NanoMQ"</code>. The result is too small for a population claim, but it is enough to show why broker management paths deserve a separate review.</p>
<p><a href="https://nanomq.io/docs/en/latest/access-control/introduction.html">NanoMQ</a> documents user-login authorization, access control lists, and HTTP authorization. Its <a href="https://nanomq.io/docs/en/latest/config-description/http-server.html">HTTP server configuration</a> includes a listening port, connection limit, authentication type, and a JWT public-key file. The <a href="https://nanomq.io/docs/en/latest/api/v4.html">HTTP API V4 documentation</a> says the API uses Basic Authentication or JWT Authentication and can query brokers, clients, and subscriptions, as well as publish messages or reload configuration.</p>
<h2>What the count covers</h2>
<p>The two ZoomEye SDK queries used <code>app="NanoMQ"</code> and <code>title="NanoMQ"</code> with <code>sub_type="all"</code>, page 1, and page size 1 at 08:42 Asia/Shanghai on 17 November 2026. The application query returned 0 and the title query returned 1. These totals describe matches for those exact fields. They do not measure MQTT listeners that do not display the product name, brokers behind a reverse proxy, or deployments reached over private networks.</p>
<p>A title-only result can be a dashboard entry page or an unauthenticated error page. It should not be called a vulnerable broker without checking the version and configuration. The same caution applies to a zero application count: this query did not find the fingerprint, and the owner still has to inspect the configured MQTT and HTTP listeners.</p>
<h2>Separate the two planes</h2>
<p>The MQTT listener carries client identity and publish/subscribe authority. Review username/password storage, JWT validation, HTTP authentication callbacks, and topic ACLs. A client that can subscribe to <code>$SYS/#</code> or a broad wildcard may learn broker and client state. A client that can publish to a wildcard or management topic may affect every device that trusts that namespace.</p>
<p>The HTTP listener can be more powerful than a dashboard. If it exposes publish, client, subscription, or configuration operations, Basic credentials and JWTs need the same secret handling as any administrative API. Bind the HTTP service to a management interface, verify the JWT key file, limit request size and rate, and log changes without recording credentials.</p>
<h2>A bounded validation path</h2>
<p>Inventory NanoMQ hosts by owner and deployment type. Check whether port 8081 or another HTTP port is reachable from the internet, whether TLS terminates in front of it, and whether Basic authentication is enabled with a unique secret. Test the ACL against a normal sensor identity and an administration identity. Confirm that disabling an account removes both MQTT and HTTP access.</p>
<p>ZoomEye gives one observable candidate in this query. The useful conclusion concerns the broker's control plane, not the size of the population.</p>
<h2>References</h2>
<ul>
<li><a href="https://nanomq.io/docs/en/latest/access-control/introduction.html">NanoMQ access control introduction</a></li>
<li><a href="https://nanomq.io/docs/en/latest/config-description/http-server.html">NanoMQ HTTP server configuration</a></li>
<li><a href="https://nanomq.io/docs/en/latest/api/v4.html">NanoMQ HTTP API V4</a></li>
</ul>
]]></content:encoded></item><item><title><![CDATA[Standard Notes separates encrypted sync from the server account boundary]]></title><description><![CDATA[Standard Notes is designed around end-to-end encryption. The applications encrypt note data before sync, and a self-hosted server provides authentication…]]></description><link>https://johnettagaweraks.hashnode.dev/standard-notes-separates-encrypted-sync-from-the-server-account-boundary</link><guid isPermaLink="true">https://johnettagaweraks.hashnode.dev/standard-notes-separates-encrypted-sync-from-the-server-account-boundary</guid><dc:creator><![CDATA[johnettagaweraks]]></dc:creator><pubDate>Tue, 17 Nov 2026 00:01:26 GMT</pubDate><content:encoded><![CDATA[<p>Standard Notes is designed around end-to-end encryption. The applications encrypt note data before sync, and a self-hosted server provides authentication and storage. That architecture reduces what a server can read from a normal note, but it does not remove account availability, authentication metadata, encryption parameters, or client trust.</p>
<p>On 2026-11-17, <code>title="Standard Notes"</code> returned 244 total matches at 07:47:12 +08:00. <code>http.body="Standard Notes"</code> returned 1,254 at 07:47:15 +08:00. Both used <code>sub_type=all</code>, page 1, and <code>pagesize=1</code>. The body population includes documentation and repositories, so it is not a server count.</p>
<h2>Self-hosting does not include the web app</h2>
<p>The official self-hosting guide says the server is the backend that processes and stores data and does not include the web application. Desktop and mobile clients connect to the server. A public host that shows documentation or a login page may be a server, a reverse proxy, or an unrelated site. The owner should record the server endpoint and the exact client configuration.</p>
<p>The encryption whitepaper describes account parameters, salts, key derivation, and MFA interactions. A malicious server can attempt to influence client cryptographic behavior, and a server that returns inconsistent parameters can cause authentication or security-update failures. Clients should be updated and pinned to trusted versions.</p>
<h2>Review the account and sync boundary</h2>
<p>Create, recover, and delete a test account. Verify the password KDF, MFA, session expiry, and server-side registration policy. Confirm that backup and file upload features have their own access controls. An end-to-end encrypted note is still unavailable if an attacker can delete the account or corrupt the sync endpoint.</p>
<p>The server operator can see connection metadata, account identifiers, session patterns, and storage volume even when note contents are encrypted. Define retention and access for those records. Recovery codes and administrative access to the mail or identity provider are part of the same recovery chain.</p>
<h2>Classify the count before naming an asset</h2>
<p>Use the 244 title matches to find candidate documentation, clients, and server pages. Verify the self-hosted server endpoint and the client application that uses it. The 1,254 body matches are useful for discovery but not proof of a live server.</p>
<p>The security value of Standard Notes comes from the client encryption and the server's authentication role together. A title count can identify where a deployment is discussed. It cannot show which server has the user's encrypted data or which client version controls the keys.</p>
<h2>References</h2>
<p>[1] ZoomEye title query. <a href="https://www.zoomeye.ai/searchResult?q=dGl0bGU9IlN0YW5kYXJkIE5vdGVzIg%3D%3D">https://www.zoomeye.ai/searchResult?q=dGl0bGU9IlN0YW5kYXJkIE5vdGVzIg%3D%3D</a></p>
<p>[2] ZoomEye body query. <a href="https://www.zoomeye.ai/searchResult?q=aHR0cC5ib2R5PSJTdGFuZGFyZCBOb3RlcyI%3D">https://www.zoomeye.ai/searchResult?q=aHR0cC5ib2R5PSJTdGFuZGFyZCBOb3RlcyI%3D</a></p>
<p>[3] Standard Notes self-hosting guide. <a href="https://standardnotes.com/help/self-hosting/getting-started">https://standardnotes.com/help/self-hosting/getting-started</a></p>
<p>[4] Standard Notes encryption whitepaper. <a href="https://standardnotes.com/help/security/encryption">https://standardnotes.com/help/security/encryption</a></p>
]]></content:encoded></item><item><title><![CDATA[CVE-2025-54236 and the CISA KEV Deadline for Magento Teams]]></title><description><![CDATA[CISA added CVE-2025-54236 to the Known Exploited Vulnerabilities catalog on October 24, 2025. For U.S. federal agencies, the listing carried a November 14…]]></description><link>https://johnettagaweraks.hashnode.dev/cve-2025-54236-and-the-cisa-kev-deadline-for-magento-teams</link><guid isPermaLink="true">https://johnettagaweraks.hashnode.dev/cve-2025-54236-and-the-cisa-kev-deadline-for-magento-teams</guid><dc:creator><![CDATA[johnettagaweraks]]></dc:creator><pubDate>Mon, 16 Nov 2026 20:41:20 GMT</pubDate><content:encoded><![CDATA[<p>CISA added CVE-2025-54236 to the Known Exploited Vulnerabilities catalog on October 24, 2025. For U.S. federal agencies, the listing carried a November 14 remediation deadline. Private-sector teams are not bound by that schedule, but the listing provides a clear signal: Adobe had already confirmed exploitation, and the flaw was no longer a theoretical patch item.</p>
<h2>Vulnerability overview</h2>
<p>Adobe APSB25-88 describes CVE-2025-54236 as an improper input validation flaw in Adobe Commerce and Magento Open Source. The CVSS 3.1 score is 9.1, with network access, low complexity, no privileges, and no user interaction. The CNA vector rates confidentiality and integrity impact as high.</p>
<p>NVD records CWE-20 and notes that the issue can enable session takeover. Sansec's research adds that the weakness can support unauthenticated RCE under certain conditions when a malicious session interacts with nested deserialization in the REST API. Adobe's own bulletin calls the impact a security feature bypass, so the two descriptions should not be merged into an unsupported claim about every target.</p>
<h2>What KEV changes</h2>
<p>A KEV entry requires affected U.S. federal civilian agencies to apply vendor mitigations or discontinue use under the applicable directive. It also gives security leaders outside government a useful prioritization reference. The date records observed exploitation.</p>
<p>The deadline does not make a vulnerable store explode on the due date. It requires an accountable remediation record. That record should show which assets were in scope, which patch or mitigation was applied, who validated the result, and whether a compromise review was needed after late remediation.</p>
<h2>Exploitation and impact</h2>
<p>Adobe confirmed active exploitation on October 22, 2025. Sansec reported that attacks began after public technical analysis and that only 38 percent of Magento stores were protected at the time. Within days, the company observed broad scanning and attempts, with PHP webshells and <code>phpinfo</code> probes in payloads.</p>
<p>A successful attacker may obtain customer-account access, modify storefront data, add administrator or integration access, and reach code execution. A store that patched late may also retain malicious session files, backdoors, or CMS modifications. The KEV response needs to include those possibilities when evidence supports them.</p>
<h2>Affected products and scope</h2>
<p>Adobe's affected Adobe Commerce versions include 2.4.9-alpha2, 2.4.8-p2, 2.4.7-p7, 2.4.6-p12, 2.4.5-p14, and 2.4.4-p15 and earlier. Adobe Commerce B2B versions 1.5.3-alpha2, 1.5.2-p2, 1.4.2-p7, 1.3.4-p14, and 1.3.3-p15 and earlier are also listed. Magento Open Source is affected through the listed 2.4.5 to 2.4.9 branches. The emergency hotfix covers 2.4.4 through 2.4.7.</p>
<h2>Exposure context</h2>
<p>ZoomEye returned 134,273 matches for <code>app="Magento"</code> and 134,273 for the selected CVE filter. The procedure selected the Search query because the counts tied. The figure describes internet-visible product fingerprints at query time and does not establish patch level or exploitation.</p>
<p>That limitation matters for KEV reporting. A broad asset count can show scale, but an agency or enterprise still needs CMDB ownership, installed version, exposure path, and remediation evidence.</p>
<h2>Remediation and mitigations</h2>
<p>Apply the Adobe hotfix or a supported security release containing the correction. Test custom modules and customer flows before production deployment. If the patch cannot meet the required schedule, document compensating controls, their coverage, and the residual risk.</p>
<p>After any delayed patch, check for webshells, anomalous sessions, new administrators, unknown API integrations, and modified CMS blocks. Sansec recommends rotating the Magento crypt key after a confirmed compromise because an exposed key can allow continued content modification. Reset administrator passwords and remove integrations that are not recognized.</p>
<p>Track closure as a verified state. The final record should include patch evidence, validation results, compromise disposition, monitoring, and an owner. A ticket marked "patched" without those checks is not enough for a known exploited vulnerability.</p>
<h2>References</h2>
<ul>
<li><a href="https://helpx.adobe.com/au/security/products/magento/apsb25-88.html">Adobe APSB25-88</a></li>
<li><a href="https://nvd.nist.gov/vuln/detail/CVE-2025-54236">NVD CVE-2025-54236</a></li>
<li><a href="https://sansec.io/research/sessionreaper">Sansec SessionReaper research</a></li>
<li><a href="https://sansec.io/research/sessionreaper-exploitation">Sansec active exploitation report</a></li>
<li><a href="https://sansec.io/research/magento-cms-block-loader-deepveindrilling">Sansec 953-store loader report</a></li>
</ul>
]]></content:encoded></item><item><title><![CDATA[Incident Response for a Suspected CVE-2026-107147 Authentication Bypass]]></title><description><![CDATA[CVE-2026-107147 creates an incident-response problem that can begin without an obvious alert. The twofactor_gauthenticator plugin for Roundcube could skip…]]></description><link>https://johnettagaweraks.hashnode.dev/incident-response-for-a-suspected-cve-2026-107147-authentication-bypass</link><guid isPermaLink="true">https://johnettagaweraks.hashnode.dev/incident-response-for-a-suspected-cve-2026-107147-authentication-bypass</guid><dc:creator><![CDATA[johnettagaweraks]]></dc:creator><pubDate>Mon, 16 Nov 2026 17:21:44 GMT</pubDate><content:encoded><![CDATA[<p>CVE-2026-107147 creates an incident-response problem that can begin without an obvious alert. The twofactor_gauthenticator plugin for Roundcube could skip its second-factor check when a password-authenticated request appeared to come from a whitelisted address. Before commit <code>4640889</code>, the plugin could source that address from <code>HTTP_CLIENT_IP</code> or a leading <code>X-Forwarded-For</code> value. A stolen password plus a forged address header could therefore produce a successful mailbox login without a valid one-time code.</p>
<p>The CVE is a CWE-290 authentication bypass with CVSS 4.0 score 6.3. The reviewed advisory sources do not report active exploitation, so a suspected case must be handled as an evidence question. Do not assume compromise from the CVE alone, and do not dismiss a suspicious login merely because the attacker presented a private address.</p>
<h2>Define the trigger</h2>
<p>A response is justified when a mailbox login looks successful without the expected second-factor step and the account or route is associated with the plugin whitelist. Other triggers include a successful password login from an unexpected network, contradictory proxy address data, or a user report of suspicious mailbox activity after a password-reset request.</p>
<p>Start by determining whether the host runs affected plugin code. The CVE describes versions from 0 before commit <code>4640889</code>. Check the plugin source or package hash. Also establish whether the whitelist feature is enabled and whether the direct peer is a proxy that supplies forwarded addresses.</p>
<h2>Preserve the evidence before cleanup</h2>
<p>Collect authentication events, proxy logs, web-server logs, deployment records, plugin configuration, and relevant mailbox audit activity. Record the direct network peer, any available forwarded chain, the user, the time, the session, and the actions that followed login. Limit access to these records because they may contain employee or customer data.</p>
<p>Capture the plugin commit and effective <code>twofactor_trusted_proxies</code> setting. If the host has already been patched, preserve the deployment record and any old image or package that identifies the prior source. Do not overwrite logs during an active investigation. If log retention is too short, document the missing interval rather than filling it with inference.</p>
<h2>Contain the exposed account and service</h2>
<p>Patch the plugin to commit <code>4640889</code> or a release that contains it. Remove unnecessary whitelist ranges and correct trusted-proxy entries. If compromise is credible, reset the affected account password, revoke active sessions, invalidate recovery material when appropriate, and review second-factor enrollment. Coordinate with the mail-platform owner before locking a production account that may be used for an active incident.</p>
<p>Isolate a web node if the address-trust configuration cannot be corrected quickly. Restrict direct internet access to Roundcube and route traffic through the intended proxy path. These changes reduce the attack surface, but they do not replace investigation of activity that may already have occurred.</p>
<h2>Investigate what the mailbox did</h2>
<p>The final login result is only the beginning of the timeline. Review new mailbox rules, forwarding settings, contacts, filters, downloads, delegated access, password-reset messages, and outbound mail. A change does not prove the CVE was exploited, but it can show whether the account produced a meaningful security consequence.</p>
<p>Correlate the mailbox timeline with password reuse or credential-leak notifications. Look for other accounts that use the same source password and for sign-in attempts against related services. Check whether the suspected login came from a proxy path that could have supplied a client-controlled header. If the evidence remains ambiguous, keep the item open as inconclusive rather than labeling it benign.</p>
<h2>Verify the repair</h2>
<p>After configuration changes, confirm that a non-whitelisted user still receives the second-factor challenge. Confirm that a legitimate whitelisted user reaches the service through every approved ingress. Test that an untrusted direct peer cannot choose the address used by the plugin. Use owned accounts only, and preserve the test evidence with the incident record.</p>
<p>ZoomEye's <code>app="Roundcube"</code> query returned 61,962 product fingerprints; the exact CVE filter returned 0. The former helps locate Roundcube assets but does not prove that the plugin or whitelist exists.</p>
<p>Close the response with a short account of what was observed, what was inferred, which accounts and hosts were affected, what was changed, and which logs remain incomplete. A suspected authentication-bypass case is resolved by evidence and repair, not by the presence of a CVE number alone.</p>
<h2>References</h2>
<ul>
<li><a href="https://cert.pl/en/posts/2026/10/CVE-2026-107145/">CERT Polska advisory</a></li>
<li><a href="https://www.cve.org/CVERecord?id=CVE-2026-107147">CVE-2026-107147 record</a></li>
<li><a href="https://github.com/alexandregz/twofactor_gauthenticator/commit/4640889d71a25391d8814de826f39416bf3db08d">Signed remediation commit</a></li>
<li><a href="https://ubuntu.com/security/CVE-2026-107147">Ubuntu CVE tracker</a></li>
</ul>
]]></content:encoded></item><item><title><![CDATA[LinkStack administration needs a public-profile and account review]]></title><description><![CDATA[LinkStack hosts a public page of links and an administrative panel for managing that page. The public side is designed to be readable by everyone. The…]]></description><link>https://johnettagaweraks.hashnode.dev/linkstack-administration-needs-a-public-profile-and-account-review</link><guid isPermaLink="true">https://johnettagaweraks.hashnode.dev/linkstack-administration-needs-a-public-profile-and-account-review</guid><dc:creator><![CDATA[johnettagaweraks]]></dc:creator><pubDate>Mon, 16 Nov 2026 14:04:53 GMT</pubDate><content:encoded><![CDATA[<p>LinkStack hosts a public page of links and an administrative panel for managing that page. The public side is designed to be readable by everyone. The control side can change identity, content, and account settings, so the two surfaces need clearly different access rules.</p>
<p>The 2026-11-15 count-only observations used page 1, pagesize 1, and <code>sub_type=all</code>. <code>app="LinkStack"</code> returned 0, <code>title="LinkStack"</code> returned 401, and <code>http.body="LinkStack"</code> returned 2,819. The body count is broad and should not be treated as a deployment count.</p>
<h2>Separate the public profile from <code>/login</code></h2>
<p>LinkStack documentation says the administrator area is reached through <code>/login</code> with the administrator credentials created by the setup wizard. Other users may register and create their own pages. Review whether public registration is intended, how accounts are verified, and who can access another user's links through the admin panel.</p>
<p>Keep the administration route on a management network when the deployment permits it. Use HTTPS and a reverse proxy that strips untrusted headers. Protect the application secrets, database, and backups because they may hold account data and content.</p>
<h2>Review account lifecycle</h2>
<p>Inventory administrators, normal users, and any hosted or multi-user configuration. Remove unused accounts, verify administrative ownership, and record the process for password reset. Confirm whether social login or other authentication methods are enabled and whether the identity mapping is narrow enough for the deployment.</p>
<p>Public profile links can point to credential reset pages, private dashboards, payment handles, or tracking URLs. Review link changes and custom HTML if that feature is enabled. A profile page change can be a brand or phishing problem even when the server remains healthy.</p>
<h2>Test the public and private sides</h2>
<p>Open the public page without a session and confirm that only intended content is visible. Try to reach the admin panel directly, then try a normal user account. Change a profile link and verify the audit or change history available to the owner.</p>
<p>For a candidate found by title, confirm the LinkStack login or public page before assigning it to an owner. The zero application result says the index did not return that exact fingerprint, not that the product is absent.</p>
<p>LinkStack is simple by design. The review stays simple when the public profile, login, registration policy, and administrator owner are recorded together.</p>
<h2>References</h2>
<ul>
<li><a href="https://docs.linkstack.org/">LinkStack documentation</a></li>
<li><a href="https://docs.linkstack.org/faq/">LinkStack FAQ</a></li>
<li><a href="https://docs.linkstack.org/setup/setup/">LinkStack setup</a></li>
<li><a href="https://github.com/LinkStackOrg/LinkStack">LinkStack repository</a></li>
</ul>
]]></content:encoded></item><item><title><![CDATA[WebAuthn credBlob stores relying-party data on the authenticator]]></title><description><![CDATA[WebAuthn credentials are usually discussed in terms of public keys, user verification, and discoverable credentials. The credBlob extension addresses a…]]></description><link>https://johnettagaweraks.hashnode.dev/webauthn-credblob-stores-relying-party-data-on-the-authenticator</link><guid isPermaLink="true">https://johnettagaweraks.hashnode.dev/webauthn-credblob-stores-relying-party-data-on-the-authenticator</guid><dc:creator><![CDATA[johnettagaweraks]]></dc:creator><pubDate>Mon, 16 Nov 2026 10:41:27 GMT</pubDate><content:encoded><![CDATA[<p>WebAuthn credentials are usually discussed in terms of public keys, user verification, and discoverable credentials. The <code>credBlob</code> extension addresses a narrower problem: a relying party may need a small amount of opaque credential-specific data available after authentication. The CTAP specification defines the authenticator behavior, while the WebAuthn specification exposes extension support and outputs.</p>
<p>The extension is not a general-purpose database. It is associated with a credential and must survive the authentication flow that returns the data. The relying party still needs a server-side record of what the blob means, which user and credential it belongs to, and whether the authenticator is trusted to return it. A blob that is accepted without a policy is just an opaque string with a familiar name.</p>
<h2>Treat the blob as a hint until verified</h2>
<p>The most useful role for <code>credBlob</code> is to carry a small non-authoritative hint, such as a credential profile identifier or a migration marker, that helps the server select a verification path. The server should verify the WebAuthn assertion, resolve the credential public key, check the user and session context, and then apply policy. The blob should not replace the credential ID, user identity, or server-side authorization record.</p>
<p>If the blob contains personal data, encryption alone does not solve retention and deletion. The relying party must document what is stored on the authenticator, how it is removed, and what happens when a credential is reset or the authenticator is lost. Data that is useful for support can also become a privacy issue when it is copied into a backup or support ticket.</p>
<h2>Authenticator support varies</h2>
<p>Extension support is a capability negotiation problem. A client or authenticator that does not support <code>credBlob</code> should fail over to a defined server-side path, not silently write a value that will never return. Registration and authentication should record whether the extension was actually used. The server must tolerate a credential that exists without a blob and a blob that exists on only one of several credentials belonging to the same account.</p>
<p>The blob is not a substitute for attestation or for the authenticator's credential protection policy. Attestation may help enrollment trust decisions, but it does not mean the blob contents are truthful. The relying party authored the value and should validate it against its own state.</p>
<h2>Plan reset and recovery</h2>
<p>A credential reset should remove or invalidate the associated server record, even if the authenticator still contains a blob. If a new credential is registered, do not assume the old blob is available. For discoverable credentials, the server should also verify that the credential it selected is the one the account policy allows.</p>
<p>Recovery must not depend on the blob being present. A user who loses an authenticator should follow the same account recovery and reauthentication controls used for any other passkey. If an administrator can copy a blob between records or manually mark it as trusted, that operation needs an audit trail and a reason.</p>
<h2>Review the complete enrollment contract</h2>
<p>Document the blob maximum size, encoding, version, creation path, return path, and deletion behavior. Test a registration without the extension, a registration with the extension, an authentication with the wrong blob, a credential reset, and an authenticator replacement. Log only the blob identifier or digest needed for troubleshooting, not the content itself.</p>
<p>The <code>credBlob</code> extension is useful when a small, credential-scoped data item genuinely needs to live on the authenticator. It becomes risky when teams treat it as a trusted claim or a convenient place to store secrets. The browser and authenticator can help return the data; the relying party remains responsible for its meaning and lifecycle.</p>
<h2>References</h2>
<ul>
<li><a href="https://www.w3.org/TR/webauthn/">Web Authentication Level 4: credBlob extension support</a></li>
<li><a href="https://fidoalliance.org/specs/fido-v2.3-rd-20251023/fido-client-to-authenticator-protocol-v2.3-rd-20251023.html">Client to Authenticator Protocol CTAP 2.3</a></li>
<li><a href="https://www.w3.org/TR/webauthn/#sctn-credBlob">WebAuthn extension overview</a></li>
</ul>
]]></content:encoded></item><item><title><![CDATA[at_hash and c_hash validation depends on one token-hash algorithm decision]]></title><description><![CDATA[OpenID Connect hybrid flows can return an ID Token from the authorization endpoint alongside an access token, an authorization code, or both. The ID Token…]]></description><link>https://johnettagaweraks.hashnode.dev/at-hash-and-c-hash-validation-depends-on-one-token-hash-algorithm-decision</link><guid isPermaLink="true">https://johnettagaweraks.hashnode.dev/at-hash-and-c-hash-validation-depends-on-one-token-hash-algorithm-decision</guid><dc:creator><![CDATA[johnettagaweraks]]></dc:creator><pubDate>Mon, 16 Nov 2026 06:43:31 GMT</pubDate><content:encoded><![CDATA[<p>OpenID Connect hybrid flows can return an ID Token from the authorization endpoint alongside an access token, an authorization code, or both. The ID Token then includes <code>at_hash</code> or <code>c_hash</code> claims so the client can bind those values to the authentication response. The check looks simple: hash the token, take the left-most half, base64url encode it, and compare. The difficult part is choosing the hash function consistently with the ID Token signing algorithm.</p>
<p>OpenID Connect Core defines <code>at_hash</code> as the base64url encoding of the left-most half of the hash of the ASCII representation of the access token. The hash uses the hash algorithm associated with the <code>alg</code> header parameter of the ID Token's JOSE header. For an <code>RS256</code> ID Token, the access token is hashed with SHA-256 and the left-most 128 bits are encoded. <code>c_hash</code> follows the same construction for the authorization code.</p>
<p>A September 2026 implementer's guide addresses algorithms whose JOSE identifier does not fully determine the digest function. That matters for implementations that support newer signature families. The guide treats <code>at_hash</code>, <code>c_hash</code>, <code>s_hash</code>, and profile-defined hash claims as one construction: select the digest and output length from the algorithm profile, hash the ASCII value, encode the required prefix, and compare the claim. The application should not choose SHA-256 merely because it is common when the ID Token uses a different hashing rule.</p>
<h2>Check the response that carried the token</h2>
<p>The claims are most useful at the authorization endpoint. A client should verify <code>c_hash</code> when the endpoint returns an authorization code, and <code>at_hash</code> when it returns an access token. The precise requirement depends on the response type and the flow. A client that ignores the claim can accept an attacker-modified token response even though the ID Token signature is valid.</p>
<p>ID Tokens returned from the token endpoint have a different context. The TLS channel protects the token exchange, and Core allows the hash claims to be omitted there even if they were present in an authorization-endpoint ID Token. Validation code should therefore know which endpoint produced the response instead of applying one generic rule to every ID Token.</p>
<h2>Make the algorithm map explicit</h2>
<p>Store the supported ID Token signing algorithms and their digest mappings in one configuration table. Reject an algorithm that is not in the table before validating the hash claim. Record the selected algorithm, digest, expected left-most length, and the result of a constant-time comparison in diagnostic logs, but never record the raw access token or authorization code.</p>
<p>Test unusual encodings as well as the normal case. The input is the ASCII representation of the token value, not the decoded JWT payload. Base64url encoding must use the unpadded form required by JOSE. A validator that compares decoded bytes without checking length can accept a truncated or extended value, while a validator that compares strings without the algorithm map can accept a value from the wrong hash family.</p>
<h2>Keep nonce, issuer, audience, and signature checks</h2>
<p>The hash claim binds the ID Token to another response value. It does not authenticate the provider, establish the client, or prevent replay by itself. Continue to verify the signature, issuer, audience, authorized party, nonce, issue time, and expiration. In a hybrid response, bind the nonce to the client's transaction and make the callback state single-use.</p>
<p>The safest implementation exposes one validation routine that knows the flow, endpoint, response type, and ID Token algorithm. It should fail closed when the token is present but the expected hash claim is missing, malformed, or computed with an unsupported algorithm. Interoperability problems should be fixed in the algorithm map, not by skipping a claim that guards the response the user is about to use.</p>
<h2>References</h2>
<ul>
<li><a href="https://openid.net/specs/openid-connect-core-1_0.html">OpenID Connect Core 1.0 incorporating errata set 2</a></li>
<li><a href="https://openid.net/specs/openid-connect-token-hash-algorithms-1_0.html">OpenID Connect Token Hash Algorithm Implementer's Guide 1.0</a></li>
<li><a href="https://openid.net/specs/openid-connect-core-1_0-36.html">OpenID Connect Core draft 36, incorporating errata set 3</a></li>
</ul>
]]></content:encoded></item><item><title><![CDATA[SparkyFitness concentrates health data behind provider and passkey settings]]></title><description><![CDATA[SparkyFitness is a self-hosted fitness, nutrition, water, and health tracker built for families. Its data model combines personal measurements with…]]></description><link>https://johnettagaweraks.hashnode.dev/sparkyfitness-concentrates-health-data-behind-provider-and-passkey-settings</link><guid isPermaLink="true">https://johnettagaweraks.hashnode.dev/sparkyfitness-concentrates-health-data-behind-provider-and-passkey-settings</guid><dc:creator><![CDATA[johnettagaweraks]]></dc:creator><pubDate>Mon, 16 Nov 2026 03:22:00 GMT</pubDate><content:encoded><![CDATA[<p>SparkyFitness is a self-hosted fitness, nutrition, water, and health tracker built for families. Its data model combines personal measurements with provider tokens and AI-assisted features. The important internet exposure question is not only who can open the dashboard, but which external services the server can reach and which credentials it can decrypt.</p>
<p>The ZoomEye count-only observation on 2026-11-16 found 0 <code>app</code> matches, 256 <code>title</code> matches, and 2,103 <code>http.body</code> matches for SparkyFitness. The probes used <code>sub_type=all</code>, page 1, and pagesize 1. The body result is a broad mention signal. The title count is the better starting set, but each candidate still needs product and ownership validation.</p>
<p>The official environment reference describes a 64-character hexadecimal <code>SPARKY_FITNESS_API_ENCRYPTION_KEY</code> used to encrypt stored external provider API keys and tokens in PostgreSQL. It also documents OIDC settings, passkey login controls, and a private-network provider switch. A deployment that changes the encryption key without a migration plan may lose access to provider credentials. A deployment that exposes the database but not the web UI still exposes the encrypted material and the rest of the health records.</p>
<h2>Separate sign-in from provider authority</h2>
<p>The web login authenticates a person to SparkyFitness. An external provider token authorizes SparkyFitness to read or write data at another service. Those are different rights. Inventory every enabled provider, its owner, its scopes, the encryption key version, and the API-key rate-limit configuration. Review whether a family member can connect a provider for another user.</p>
<p>Passkey registration needs an existing session. The environment guide warns that a fresh instance with neither OIDC nor SMTP-backed magic links can leave no way to sign in, because passkey registration depends on an authenticated user. Plan the bootstrap account and recovery path before disabling local login. Keep administrator recovery separate from normal family access.</p>
<h2>Treat health records as sensitive production data</h2>
<p>Back up the PostgreSQL database and the encryption key under separate controls. A backup without the key is unreadable, and a key backup with the database defeats encryption. Test restoration in an isolated environment. Restrict database network access and remove public object-storage or provider callback routes that are no longer needed.</p>
<p>Private-network provider access deserves particular caution. A feature that lets the server reach a self-hosted recipe or food provider can also become a server-side request path. Allow only configured providers and networks, block link-local and metadata addresses, and log the destination without storing returned health data. The same review applies to AI providers and any integration that receives family records.</p>
<h2>Build an ownership record</h2>
<p>For each title match, verify the application assets, login page, registration state, OIDC provider, API version, and version evidence. Ask the owner whether the instance is personal, shared with a household, or intended for public access. Record the result and observation time.</p>
<p>SparkyFitness can hold enough information to reveal routines, location, diet, and health conditions. The 256-title result should be treated as a candidate list for owner review. A dashboard behind a password is a good start. The provider tokens, database, encryption key, backups, and recovery paths determine whether the data stays inside that boundary.</p>
<h2>References</h2>
<ul>
<li><a href="https://github.com/CodeWithCJ/SparkyFitness">SparkyFitness repository</a></li>
<li><a href="https://codewithcj.github.io/SparkyFitness/install/environment-variables">SparkyFitness environment variables reference</a></li>
<li><a href="https://github.com/CodeWithCJ/SparkyFitness/blob/main/docker/docker-compose.prod.yml">SparkyFitness production compose file</a></li>
<li><a href="https://www.zoomeye.ai/searchResult?q=dGl0bGU9IlNwYXJreUZpdG5lc3Mi">ZoomEye count-only observation</a></li>
</ul>
]]></content:encoded></item><item><title><![CDATA[Apple TV Is a Network Endpoint: Segmenting CVE-2026-112687 Exposure]]></title><description><![CDATA[Apple TV devices often disappear from security inventory because they do not look like traditional computers. They connect to the network, run an operating…]]></description><link>https://johnettagaweraks.hashnode.dev/apple-tv-is-a-network-endpoint-segmenting-cve-2026-112687-exposure</link><guid isPermaLink="true">https://johnettagaweraks.hashnode.dev/apple-tv-is-a-network-endpoint-segmenting-cve-2026-112687-exposure</guid><dc:creator><![CDATA[johnettagaweraks]]></dc:creator><pubDate>Sun, 15 Nov 2026 17:22:08 GMT</pubDate><content:encoded><![CDATA[<p>Apple TV devices often disappear from security inventory because they do not look like traditional computers. They connect to the network, run an operating system, receive updates, and interact with media services. CVE-2026-112687 makes that endpoint role relevant: Apple says a local-network attacker can cause system termination or corrupt process memory through the MediaRemote component on Apple TV 4K second generation and later.</p>
<p>The tvOS fix is 27.2. The same CVE also affects specified iPhone and iPad models through the iOS and iPadOS 27.2 updates. The Apple TV portion deserves separate attention because ownership, update timing, and network placement differ from phones and tablets.</p>
<h2>Find the hidden estate</h2>
<p>The affected tvOS hardware begins with Apple TV 4K second generation. A facilities team may know where the devices are installed, while information technology knows which network jacks they use. Neither group may own the update process.</p>
<p>Build an inventory from purchase records, wireless controllers, DHCP logs, DNS observations, and switch ports. Match physical location to a business function and a management path. A display in a reception area, a conference room, and a production control room can share the same model while carrying different consequences.</p>
<p>Do not assume that an Apple TV is unimportant because it has no endpoint agent. Its network position may provide a path to other systems or a visible service used by customers and staff.</p>
<h2>Understand the MediaRemote boundary</h2>
<p>Apple attributes the flaw to MediaRemote and states that an attacker on the local network can cause the documented impact. The advisory does not publish the exact trigger or a public proof of concept. Network planning must therefore use the known attack position and impact rather than a guessed protocol rule.</p>
<p>The remediation removes the vulnerable code. That is a vendor-side fix. Network segmentation is a temporary control that reduces who can reach a vulnerable endpoint. It should be documented as containment, not as a replacement for 27.2.</p>
<p>A media device may need to communicate with phones, tablets, displays, or controllers. If those flows are required, restrict them to known segments and systems. If a device only needs outbound access for updates and streaming, the broad local-network path may be unnecessary.</p>
<h2>Review peer-to-peer paths</h2>
<p>Local media coordination can involve discovery and direct connections between devices. The public advisory does not identify which traffic reaches CVE-2026-112687, so a team cannot block a single published signature. It can still review the environments where such coordination occurs.</p>
<p>Check whether guest wireless clients can reach the Apple TV segment. Review multicast and broadcast boundaries where the network design exposes them. Confirm that conference-room devices are not reachable from unmanaged user VLANs unless a business workflow requires it.</p>
<p>Change control matters. Blocking discovery or direct media traffic can interrupt casting, remote controls, or room automation. Exercise the narrowest restriction that preserves the approved workflow, then document the exception and the review date.</p>
<h2>Patch the device, not only the dashboard</h2>
<p>A device record can remain in a management system after the hardware is replaced or powered off. Verify the current tvOS version, the last check-in, and the physical device identity. Apple's fixed release is tvOS 27.2 for Apple TV 4K second generation and later.</p>
<p>Some media endpoints may be difficult to update because no one knows the local account or because the device is mounted behind a display. That operational detail belongs in the exception record. A team can plan a site visit or temporary network restriction until the update is complete.</p>
<p>If a device crashes or restarts unexpectedly, preserve the timing and current network peers before replacing it. The Apple advisory names system termination and process-memory corruption, but it does not provide an exploit detection rule. A reboot is a symptom to investigate, not proof of compromise.</p>
<h2>Use exposure data for the right question</h2>
<p>ZoomEye's <code>app="iOS"</code> query returned 1,591,950 product-fingerprint matches during this review. The exact <code>vul.cve="CVE-2026-112687"</code> query returned zero. These figures do not enumerate internal Apple TVs, and the product query is not a vulnerable-device count.</p>
<p>External observations can help locate internet-facing assets or understand the scale of indexed Apple technology. The local-network condition in this CVE points toward internal discovery and segmentation instead. A complete program uses both views without turning either into a claim of confirmed compromise.</p>
<h2>Define a safe media network</h2>
<p>A defensible design gives each media endpoint the connectivity it needs and no broader peer population. Document the source systems, protocols, and support owner. Test the update and media workflow after each firewall or wireless-isolation change. Retain logs long enough to investigate an unusual crash or connection during the change window.</p>
<p>Completion has two parts. Every supported Apple TV should run tvOS 27.2 or have an approved exception, and each exception should have a documented network restriction and an owner. The media network is then easier to defend, inventory, and update when the next local flaw affects it.</p>
<h2>References</h2>
<ul>
<li><a href="https://support.apple.com/en-us/149524">Apple: About the security content of tvOS 27.2</a></li>
<li><a href="https://support.apple.com/en-us/149230">Apple: About the security content of iOS 27.2 and iPadOS 27.2</a></li>
<li><a href="https://advisories.ncsc.nl/2026/ncsc-2026-0456.html">NCSC-NL: Security Advisory NCSC-2026-0456</a></li>
<li><a href="https://nvd.nist.gov/vuln/detail/CVE-2026-112687">NVD: CVE-2026-112687</a></li>
</ul>
]]></content:encoded></item><item><title><![CDATA[FastGPT's 3,306 Application Fingerprints Point to a Team and Data Authority Review]]></title><description><![CDATA[FastGPT is a self-hostable knowledge-base and application platform. A ZoomEye count-only probe on November 15, 2026 returned 3,306 app="FastGPT" matches…]]></description><link>https://johnettagaweraks.hashnode.dev/fastgpt-s-3-306-application-fingerprints-point-to-a-team-and-data-authority-review</link><guid isPermaLink="true">https://johnettagaweraks.hashnode.dev/fastgpt-s-3-306-application-fingerprints-point-to-a-team-and-data-authority-review</guid><dc:creator><![CDATA[johnettagaweraks]]></dc:creator><pubDate>Sun, 15 Nov 2026 14:02:27 GMT</pubDate><content:encoded><![CDATA[<p>FastGPT is a self-hostable knowledge-base and application platform. A ZoomEye count-only probe on November 15, 2026 returned 3,306 <code>app="FastGPT"</code> matches, 310 <code>title="FastGPT"</code> matches, and 11,198 <code>http.body="FastGPT"</code> matches. The scope was <code>sub_type=all</code>, page 1, and page size 1.</p>
<p>The application count is the strongest of the three product-specific observations, but it is still not a count of vulnerable or publicly usable deployments. A fingerprint can identify an application surface without proving that a user can log in, that a team is active, or that a dataset contains sensitive material. The count is useful for prioritization and follow-up.</p>
<h2>The API boundary is layered</h2>
<p>FastGPT's OpenAPI documentation separates application access from team and resource permissions. An application key can invoke a published app, while the application itself may rely on team membership, a dataset, and configured tools. The backend that receives a request still needs to authenticate the caller and decide whether that caller may use the app.</p>
<p>Do not treat an app key as a workspace administrator credential. Inventory which user or team owns the key, which application ID it targets, and whether the key can use development or system endpoints. The official integration guidance distinguishes these API surfaces and says to check the authentication method for each endpoint.</p>
<p>If an OpenAI-compatible endpoint is enabled for convenience, verify its app ID resolution and key format. A compatibility layer that accepts the wrong key shape can attach a request to an unintended app. Test with a key from a different team and confirm that the request is rejected.</p>
<h2>Datasets and tools carry authority</h2>
<p>A knowledge-base application is valuable because it connects model calls to retrieved data and tools. Review dataset membership, document ingestion, vector-store access, model-provider keys, and HTTP tool credentials. A public chat interface that uses a private dataset can expose content without exposing the dataset administration page.</p>
<p>Limit server-side HTTP tools to approved destinations. A tool that can call arbitrary URLs creates an egress boundary even when the chat user cannot reach the target. Keep model provider keys in the server's secret store and rotate them if an application or workspace is transferred.</p>
<p>Team membership is not just a UI feature. It decides who can see applications, datasets, and usage. Remove inactive members and service accounts, and confirm that removing a member invalidates sessions and API keys that were tied to that identity.</p>
<h2>Review the discovered instances</h2>
<p>Start with application fingerprints that return a FastGPT interface. Verify the deployment type and whether it is public, internal, or a demo. For each confirmed instance, record the team model, public app inventory, signup policy, API keys, and dataset owners.</p>
<p>The 310 title matches and 11,198 body matches capture different evidence. A title match may be a branded login page. A body match may be a help article or an integration page. Keep the exact dorks and observation time with the finding, and do not sum the categories as if they were unique hosts.</p>
<p>Monitor for changes in the application fingerprint around known deployment ranges. A new instance may be a test environment with copied data. A disappearing fingerprint may mean a host was removed or that the response changed, not necessarily that the application is gone.</p>
<h2>Measure the control plane</h2>
<p>The practical inventory fields are owner, deployment URL, authentication method, team count, public app count, dataset count, model-provider account, tool destinations, and API-key owners. Add a review date and the last verified configuration version.</p>
<p>FastGPT concentrates several sensitive relationships in one interface. The ZoomEye fingerprint is a way to find candidate control planes. The security decision is whether the team, dataset, tool, and key boundaries match the data the application can reach.</p>
<h2>References</h2>
<ul>
<li><a href="https://doc.fastgpt.io/en/openapi/intro">FastGPT OpenAPI introduction</a></li>
<li><a href="https://doc.fastgpt.io/en/guide/build/publish/openapi">FastGPT app API access</a></li>
<li><a href="https://www.zoomeye.ai/">ZoomEye count evidence for this article</a></li>
</ul>
]]></content:encoded></item></channel></rss>