Feast through 0.66.0 fails to verify JWT token signatures before establishing user identity, allowing attackers to bypass all role-based access control by presenting an unverified token with a hardcoded claim value. Attackers can obtain trusted internal identity and gain unchecked read and write access to all entities, feature views, data sources, and permission policies on the server.

LightGBM through 4.7.0 fails to validate child and split array values when parsing text models, allowing attackers to write out-of-bounds memory during SHAP prediction. Attackers can craft malicious model files with invalid node references that trigger out-of-bounds writes at attacker-chosen offsets in the leaf_depth_ buffer during feature contribution computation.

Angel through 3.3.0 deserializes untrusted setAlgoMetrics payload using Kryo without class registration or allowlist validation. Unauthenticated network attackers can instantiate arbitrary classes or exhaust coordinator memory by sending crafted serialized objects to the master RPC endpoint.

@refinedev/inferencer through 7.0.0 fails to escape API field names when interpolating them into generated JSX source code. Attackers controlling the data provider can inject malicious JavaScript through crafted JSON property names that execute in the developer's browser when the Inferencer page renders.

Yeti through 2.11.0 fails to validate caller permissions in the DELETE /api/v2/rbac/{id} endpoint, allowing users with read access to delete access control relationships. Attackers can revoke the owner's grant and permanently lock legitimate owners out of objects.

Chroma through 1.5.9 fails to validate tenant and database segments when resolving collections, allowing authenticated attackers to access collections from other tenants by knowing the collection identifier. Attackers can read, modify, and update records in foreign collections by issuing requests under their own tenant path, bypassing authorization checks.

Builder.io Gen2 SDKs through versions 5.2.11 and 0.25.13 contain a prototype pollution vulnerability in the unflatten helper that processes builder.userAttributes query parameters without prototype guards. Attackers can craft preview links with __proto__ or prototype segments to pollute Object.prototype in a visitor's browser when the SDK processes the malicious URL.

KnowStreaming through 3.4.1 fails to enforce role-based access control on REST API endpoints, allowing any authenticated user to access protected functionality. Attackers can call identity-management endpoints to create administrator accounts or grant themselves administrative privileges without proper authorization.

shell-quote's `quote()` function did not validate object-token inputs against the operator model used by `parse()`. The `.op` field was backslash-escaped character by character using `/(.)/g`, which in JavaScript does not match line terminators (\n, \r, U+2028, U+2029). A line terminator in `.op` therefore passed through unescaped into the output; POSIX shells treat a literal newline as a command separator, so any content after it would execute as a second command. The vulnerable code path is reachable in two ways: (1) direct construction of `{ op: '...\n...' }` from external input, and (2) via `parse(cmd, envFn)` when `envFn` returns object tokens whose `.op` is attacker-influenced. Both are documented API surface. Fixed by replacing the per-character escape with strict shape validation: `.op` must match the parser's control-operator allowlist; `{ op: 'glob', pattern }` validates `pattern` and forbids line terminators; `{ comment }` validates `comment` and forbids line terminators; an

Builder.io Gen2 SDKs through versions 5.2.11 and 0.25.13 contain a prototype pollution vulnerability in the deep-set helper function that processes content block bindings without validation. Attackers can craft content blocks with binding keys containing __proto__, prototype, or constructor paths to pollute Object.prototype during rendering, affecting all subsequent objects created in the process including other tenants' renders.

CMAK through 3.0.0.6 fails to apply the scheduled leader election feature toggle to HTML form routes, allowing attackers to bypass the feature gate. Attackers can access the form endpoints to start and stop the recurring election scheduler, disrupting leadership across managed Kafka clusters.

Wiki.js through 2.5.314 fails to require path separators when matching START and END page rules, allowing attackers to access pages sharing a prefix with authorized folders. Users granted access to a folder can read and modify unrelated pages with matching prefixes, bypassing intended access controls.

Wiki.js through 2.5.314 contains a server-side request forgery vulnerability in the Image Prefetch renderer that fetches arbitrary URLs without protocol, host, or address validation. Attackers with page editing permissions can inject img elements with the prefetch-candidate class to make the server request internal services and cloud metadata endpoints, with responses returned to the attacker.

Wiki.js through 2.5.314 omits page tags from authorization checks in multiple GraphQL resolvers, allowing tag-based access restrictions to be bypassed. Attackers can query the list, tree, tags, searchTags, and links resolvers to retrieve restricted page metadata including titles, descriptions, paths, and tag information without proper authorization.

Trigger.dev before 4.6.0 fails to verify that an authenticated user controls a GitHub App installation before binding it to their organization. Attackers can claim another user's GitHub App installation by replaying state cookies and supplying sequential installation identifiers, gaining unauthorized access to the victim's repositories.

Leantime before 3.9.6 contains an authorization bypass vulnerability in the HTMX plugin install endpoint that lacks permission validation. Authenticated users with limited roles can install marketplace plugins and control arbitrary properties including identifier, version, and license key to deploy malicious plugins.

Twenty before 2.35.0 fails to validate field and row permissions in the groupBy-with-records GraphQL resolver, allowing authenticated users to bypass permission checks. Attackers with canReadObjectRecords permission but canReadFieldValue false can retrieve restricted field values through the groupBy resolver that would normally be denied.

Harbor through 2.15.2 fails to properly restrict the q query parameter filtering on scanner registration access credentials. Project administrators can exploit fuzzy filtering on the AccessCredential column to recover the scanner adapter secret one character at a time through response row counts.

ArcherySec through 2.0.6 fails to validate organization ownership in the WebScanVulnList endpoint, allowing authenticated users to read vulnerability findings from other organizations. Attackers can supply arbitrary scan identifiers to retrieve complete web vulnerability data including titles, severities, statuses, and analyst notes from other tenants.

OpenCVE before 3.1.0 fails to properly scope the organizations API endpoint to the token's organization, instead returning the token creator's memberships. Attackers with organization-scoped tokens can list and retrieve every organization their creator belongs to, bypassing intended token isolation boundaries.