CVE-2026-118668
langgraph-api implements the LangGraph API for rapid development and testing. From 0.1.2 until 0.15.0, JavaScript or TypeScript custom authe
CVSS
—
No CVSS
EPSS
—
KEV
—
Exploit Today
—
0-100
Published: Oct 28, 2026 · Last modified: Oct 28, 2026 · CWE-639 · CWE-863
Not enough EPSS history yet.
langgraph-api implements the LangGraph API for rapid development and testing. From 0.1.2 until 0.15.0, JavaScript or TypeScript custom authentication Store authorization handlers can return a rewritten namespace that is not carried across the runtime boundary and applied before the Store operation. An authenticated user can therefore access another user's Store namespace when a deployment relies on namespace rewriting for isolation, allowing unauthorized reads, overwrites, searches, or deletions. Python custom authentication and handlers that reject unauthorized namespaces instead of rewriting them are not affected. This issue is fixed in version 0.15.0.
CVECVSSEPSSKEVRExploitTitleMod.
CVE-2026-857498.1 HIG—
———MailWatch for MailScanner is a web-based front-end to MailScanner. Prior to 1.2.27, any authenticated MailWatch user, including the U/user, D/domain-admin, and H/host roles, can release to inbox or permanently delete the quarantined mail of any other user or domain by supplying an arbitrary message `id` to `mailscanner/quarantine_action.php`. The action path calls `quarantine_list_items()` in `mailscanner/functions.php`, which selects from `maillog` by the supplied identifier without the per-role `global_filter` applied by listing and read views. The action requires a valid session and CSRF token, but the caller can submit a token from the caller's own session; ownership of the requested message is never checked. Message identifiers are enumerable, allowing a low-privilege user to invoke `quarantine_release()` or `quarantine_delete()` for other users' or domains' quarantined messages, force-delivering mail held by the quarantine or permanently deleting it. This issue is fixed in version 1.2.27.2hCVE-2026-824116.5 MED—
———Open Object provides an API and admin interface to manage objects and object types. From 2.4.3 until 2.5.1, 3.6.2, 4.0.2, and 4.1.1, a token scoped to read one ObjectType can read the full versioned record of an object of any other ObjectType via the history-detail endpoint `GET /api/v2/objects/{uuid}/{index}`. The `history_detail` action in `src/objects/api/v2/views.py` calls `self.get_queryset()` and `get_object_or_404` directly instead of `self.get_object()`, so `check_object_permissions` and the per-object check in `ObjectTypeBasedPermission.has_object_permission` are not invoked. The queryset's `filter_for_token` applies only to the list and search actions, while `ObjectTypeBasedPermission.has_permission` allows non-create actions, leaving neither check to constrain history-detail records to the token's permitted ObjectType. The normal retrieve endpoint returns 403 for the same unauthorized object, and the sibling history action enforces the grant. Any valid non-superuser API token with a read grant on at least one ObjectType can therefore disclose another ObjectType's complete versioned object `data`, which can include citizen or case records and sensitive identifiers in government deployments. This issue is fixed in versions 2.5.1, 3.6.2, 4.0.2, and 4.1.1.2hCVE-2026-55561——
———InvoicePlane is a self-hosted open source application for managing invoices, clients, and payments. Prior to 1.7.2-beta-1, guest users can approve or reject quotes belonging to any client, not just their assigned clients. The `approve()` and `reject()` methods in the Guest Quotes controller, `application/modules/guest/controllers/Quotes.php`, call `approve_quote_by_id($quote_id)` and `reject_quote_by_id($quote_id)` directly without the `where_in('ip_quotes.client_id', $this->user_clients)` scoping that sibling methods use. A guest user assigned to one client can change another client's quote status by supplying sequential integer `quote_id` values, and `email_quote_status()` also sends notification emails. Additionally, `generate_pdf()` calls `mark_viewed($quote_id)` before checking the client scope, allowing a guest user to mark another client's quote as viewed. These issues are fixed in version 1.7.2-beta-1 and the subsequent stable version 1.7.2.2hCVE-2026-118669——
———OAuthLib is an implementation of the OAuth request-signing logic for Python 3.6+. Prior to 4.0.0, the client_credentials and Resource Owner Password Credentials token grant handlers can authenticate request.client as one confidential client while retaining an attacker-supplied request.client_id for validate_grant_type and validate_scopes. A lower-privilege authenticated client can therefore cause grant and scope authorization to use another client's identity when an application validator relies on the documented client_id argument, potentially issuing tokens with unauthorized grants or scopes. Validators that authorize exclusively from request.client may not be affected. This issue is fixed in version 4.0.0.3hCVE-2026-1186507.5 HIG—
———Echo is a Go web framework. Prior to 4.16.0 and 5.4.0, a globally mounted middleware.Static instance can interpret an encoded request path differently from Echo's router because the router matches the still-encoded path while the static handler uses the decoded URL.Path. The advisory demonstrates an encoded slash resolving to a file under the static root through the global static middleware, but the source is internally inconsistent about whether a route or group guard protects the corresponding literal path in that global-mount configuration. This issue is fixed in 4.16.0 and 5.4.0.5hCVE-2026-1175749.1 CRI—
———A flaw was found in ai-gateway-payload-processing. An authenticated remote attacker can exploit this vulnerability by submitting a request specifying another tenant's model identifier in the request body or header. Because the service does not verify authorization before retrieving credentials from the shared store, it injects the target tenant's application programming interface (API) keys into the request. This flaw allows an attacker to perform unauthorized requests and consume another tenant's large language model (LLM) resources and quotas.5h