CVE Feed

vLLM through 0.29.0 fetches and fully materializes remote or inline media before enforcing its documented media controls (the VLLM_MAX_AUDIO_CLIP_FILESIZE_MB compressed-audio size cap, default 25 MB, and the per-modality --limit-mm-per-prompt item limits). Across four ingress paths — the shared media-acquisition layer (HTTPConnection.get_bytes()/async_get_bytes()), the chat completions audio_url/base64 path, the batch speech runner, and the Rust frontend POST /tokenize route — the server reads the entire HTTP response body, base64-decodes the inline payload, or spawns one fetch/decode task per media part, and only then applies the limit (or, on some paths, never applies it). A remote attacker can therefore cause the API server or batch-runner process to allocate memory and consume outbound bandwidth proportional to an attacker-chosen body size or media item count before the request is rejected, resulting in pre-inference memory and bandwidth exhaustion (denial of service). The chat and batch surfaces require an API key when one is configured; the Rust frontend /tokenize route is unauthenticated by design. There is no code execution or data disclosure impact.

vLLM before 0.29.0 contains a resource-limit bypass vulnerability in PyNvVideoCodec decoder allocation where sampler subclass shadowing allows independent counter increments. Unauthenticated attackers can select different sampler subclasses in video requests to exceed configured decoder limits and exhaust unaccounted GPU memory.

vLLM versions before 0.29.0 contain a denial-of-service vulnerability in the cache_salt parameter accepted on OpenAI-compatible and Anthropic API endpoints, which lacks maximum length validation and is processed on the single EngineCore scheduler thread. Unauthenticated attackers can send HTTP requests with multi-hundred-megabyte salt values that trigger expensive pickle serialization and SHA-256 hashing, stalling the scheduler thread and denying service to all concurrent requests.

vllm before 0.29.0 fails to enforce VLLM_MAX_AUDIO_CLIP_FILESIZE_MB limit in multimodal chat audio decoding, allowing unauthenticated clients to bypass file size restrictions. Attackers can submit oversized audio files through chat endpoints to consume excessive memory and CPU resources during decoding.

SiYuan is a self-hosted personal knowledge management system. In versions up to and including 3.8.3, the kernel's authentication guards (CheckAuth in kernel/model/session.go and IsSessionOriginAllowed in kernel/util/net.go) fail open when the HTTP Origin header is absent, on the incorrect assumption that any browser-initiated cross-site request carries an Origin. Because browsers omit Origin on cross-site top-level GET navigations and no-cors GET subresource loads — and the session cookie is SameSite=Lax — a single cross-site GET issued from any attacker-controlled web page is granted RoleAdministrator, both on default installations with no access-authorization code and on password-protected instances with a live session. Combined with content-type sniffing on the /api/network/proxy endpoint, which allows attacker-controlled HTML to be served under SiYuan's own origin, this permits an unauthenticated remote attacker to execute arbitrary script in the SiYuan origin (http://127.0.0.1:6806), invoke administrator APIs, and exfiltrate the persistent kernel API token. This issue is fixed in version 3.8.4.

SiYuan versions 3.7.0 before 3.8.4 contain a stored cross-site scripting vulnerability in gallery and kanban database renderers where field descriptions are not escaped in aria-label attributes. In the Electron desktop app with nodeIntegration enabled, attackers can inject JavaScript that calls Node.js child_process APIs to execute arbitrary commands with user privileges.

SiYuan before v3.8.4 contains a SQL injection vulnerability in the graph query endpoint where the dailyNoteSavePath parameter is concatenated into SQL without escaping. Unauthenticated attackers on published sites with auth disabled can inject SQL via UNION SELECT to extract arbitrary database rows from all notebooks.

SiYuan versions before v3.8.4 fail to properly escape four stored Attribute View values in textarea elements, allowing authenticated attackers to inject JavaScript by modifying field descriptions, template sources, select option descriptions, or footer calculation templates. Attackers can execute stored JavaScript when other users open affected database menus, and in the Electron desktop app with nodeIntegration enabled, this leads to command execution with SiYuan process privileges.

SiYuan versions from v2.1.0 before v3.8.4 contain a cross-site request forgery vulnerability in the CheckAuth lock-screen pass-through branch that grants administrator access to loopback requests without validating Origin headers. Attackers can craft malicious web pages that force victims to terminate the kernel process, read workspace configuration and proxy settings, and trigger administrative actions via zero-credential cross-origin requests from the victim's browser.

SiYuan before v3.8.4 does not HTML-escape stored flashcard block content before interpolating it into the card-manager list markup. Block content returned by /api/riff/getRiffCards is inserted into a card item template in app/src/card/viewCards.ts and assigned to listElement.innerHTML, so content such as <img src=invalid onerror=...> becomes an executable event-handler attribute. Because the SiYuan desktop (Electron) main window is created with nodeIntegration enabled and contextIsolation disabled, an administrator who opens the card manager on a workspace containing attacker-supplied flashcard content (for example introduced through contribution or import) executes the attacker's script in a privileged renderer, which can lead to arbitrary code execution on the host. The affected endpoint remains behind authentication and administrator-role checks; this is an untrusted-content-to-privileged-renderer issue, not an authorization bypass.

SiYuan before v3.8.4 contains an authorization omission in the siyuan-get IPC handler that allows remote-kernel renderers to access native clipboard formats by invoking clipboardReadMathML, clipboardReadOffice, and clipboardReadWPS commands with matching plaintext. Attackers controlling remote renderer content can obtain MathML formulas, Office bytes, and WPS bytes from local clipboard during user-mediated paste operations.

SiYuan v3.8.3 fails to HTML-escape the data-subtype attribute when generating gutter-button markup (app/src/protyle/gutter/button.ts, assigned via innerHTML in app/src/protyle/gutter/index.ts) from content pasted as plain-text Markdown containing a Kramdown inline attribute list (IAL). Because the shared Lute renderer parses Kramdown IAL from text/plain input, an attacker-supplied Markdown snippet using entity-encoded quotes in data-subtype breaks out of the attribute value when the gutter markup is re-parsed by the browser, injecting additional attributes such as autofocus and onfocus. If a victim pastes the crafted Markdown and the affected gutter control receives focus, the injected handler executes; in the Electron desktop application, where the main BrowserWindow enables Node integration and disables context isolation, this results in JavaScript execution with renderer Node.js privileges (remote code execution). Fixed in v3.8.4.

SiYuan versions before v3.8.4 contain a path traversal vulnerability in the setNotebookIcon endpoint that allows authenticated administrators to create arbitrary directory trees and write files outside the workspace boundary. Attackers can supply directory traversal sequences in the notebook parameter to escape the workspace data directory and write conf.json files to arbitrary locations accessible by the kernel process.

SiYuan versions before v3.8.4 contain a path traversal vulnerability in the checkoutRepo endpoint that allows authenticated administrators to write JSON files outside the workspace. Attackers can supply a sessionID parameter containing directory traversal sequences to overwrite arbitrary JSON files in pre-existing kernel-writable directories outside workspace boundaries.

SiYuan versions before v3.8.4 contain a path traversal vulnerability in the exportBrowserHTML endpoint that allows authenticated administrators to write arbitrary HTML content to index.html outside the workspace directory. Attackers can supply a folder parameter with directory traversal sequences to escape the export directory and overwrite index.html in any pre-existing kernel-writable location, enabling stored XSS or workspace defacement.

SiYuan before v3.8.4 contains an authentication bypass vulnerability in the publish service where session cookies are issued without Secure or SameSite attributes over plaintext HTTP connections. An on-path attacker can observe a valid publish-visitor-session-id cookie from a Basic Auth exchange and replay it to access authenticated publish endpoints without knowing the account password.

SiYuan before v3.8.4 does not validate the sender or restrict recipients in the 'siyuan-send-windows' IPC handler of the Electron main process (app/electron/main.js). The handler ignores event.sender and forwards any received payload to every BrowserWindow returned by BrowserWindow.getAllWindows(), including windows belonging to other opened workspaces. A renderer connected to an attacker-controlled remote kernel can therefore send {cmd: "lockscreenByMode"} and have it delivered across the workspace boundary; a sibling workspace window whose lockScreenMode is set to 1 invokes lockScreen(). Repeated messages allow the remote workspace to repeatedly lock unrelated local workspace windows, causing a limited denial of service. No confidentiality, integrity, or code-execution impact was observed.

SiYuan is a self-hosted personal knowledge management system. In versions 3.8.0 through 3.8.3, the MCP file tool's sensitive-path guard (util.IsForbiddenAbsPath(), invoked from resolvePath()) is applied only to the allowed root of recursive operations and not to each resolved descendant path — an incomplete fix for GHSA-c8r8-95hg-mp34. An authenticated administrator using the in-app Agent or the external MCP server can therefore bypass the protected-workspace-file denylist: file.grep can return matching lines from non-hidden protected descendants (for example conf/conf.json, TLS keys, data/snippets/conf.json, data/templates/, data/.siyuan/publishAccess.json, notebook .siyuan internals, or the kernel log), file.copy can copy protected descendants to an ordinary path where file.read can then retrieve them, and unzip can overwrite protected descendants using ordinary, lexically contained ZIP member names. Because file.grep is globally classified as a safe action, it receives no per-call confirmation, and the confirmation cards for file.copy and unzip show only the allowed root arguments. This issue is fixed in version 3.8.4. Suggested title: SiYuan 3.8.0 through 3.8.3 Sensitive-Path Guard Bypass in Recursive MCP File Operations.

Parse Server is an open-source backend server. In versions >= 9.0.0 and < 9.10.1-alpha.8, and in versions < 8.6.89, LiveQuery evaluates the protectedFields class-level permission against an incompletely resolved caller identity: the subscriber's roles are not resolved, and when a subscription does not supply its own session token the event payload is redacted against an anonymous identity even though the read was authorized against the connected user. As a result, field masks defined for a role, for authenticated users, or for a specific user are not applied, so an authenticated subscriber can receive field values that the REST API correctly withholds and can use a masked field to filter or watch a subscription. Only classes with LiveQuery enabled that define protectedFields under a role:, authenticated, or per-user group are affected; masks under the public (*) group are applied correctly. The issue is fixed in 9.10.1-alpha.8 and 8.6.89. As a workaround, additionally define the affected field masks under the public (*) group, or disable LiveQuery for classes whose class-level permissions rely on role-scoped, authenticated, or per-user protectedFields groups.

Parse Server is an open source backend server. In versions prior to 8.6.90 and in versions from 9.0.0 prior to 9.10.1-alpha.9, the device token deduplication logic for installation records does not validate the type of client-supplied installation fields before using them to build database queries. An unauthenticated remote attacker who knows only the public application ID can submit non-string values in these fields to inject query operators, causing the deduplication cleanup — which runs with elevated privileges before class-level permissions are evaluated — to delete every device registration in the application or an attacker-chosen subset of them. No account, session token, master key, or user interaction is required. Deleted registrations cannot be recovered on the server, so push notifications cannot be delivered until every client re-registers. Any deployment that exposes the REST API to clients and uses push notifications is affected in its default configuration. Versions 8.6.90 and 9.10.1-alpha.9 fix the issue by rejecting non-string values with a client error and by scoping the deduplication cleanup to the calling application. No workaround other than upgrading is available.

AVideo before 29.1.0 contains a stored cross-site scripting vulnerability in the video trailer1 field rendered unsanitized within an inline onclick JavaScript string. Attackers with video upload permission can store HTML entity-encoded payloads that bypass isValidURL() validation and are decoded by the browser to break out of the JavaScript string, executing arbitrary code in any visitor's session including administrators.

capgo.app before 12.128.12 fails to enforce an organization's API key expiration policy when creating app-scoped API keys. In the POST /apikey endpoint, requests that supply app_id but omit org_id, limited_to_orgs, and expires_at resolve the target app and scope the key to it, but never add the app's owner organization to the list of organization IDs passed to validateExpirationAgainstOrgPolicies; because that list is empty, the validation returns early. As a result, an authenticated organization member can create a non-expiring app-scoped API key even when the owning organization has require_apikey_expiration enabled and a max_apikey_expiration_days limit configured. The issue is fixed in version 12.128.12.

Capgo (capgo.app backend) before 12.127.5 contains an authorization flaw in the PATCH /private/role_bindings/:binding_id endpoint. The handler verifies that the newly assigned role's priority rank does not exceed the caller's own rank, but — unlike the DELETE handler — it never checks the rank of the role currently bound to the target binding. An authenticated user holding the org_admin role (rank 90) can therefore change an org_super_admin binding (rank 95) to a lower-privileged role such as org_member (rank 75). Because the prevent_last_super_admin_binding_delete database trigger fires only BEFORE DELETE and not on UPDATE, an org_admin can demote every org_super_admin, leaving the organization with no super administrator. The issue is fixed in 12.127.5.

Capgo (Cap-go/capgo.app) server backend Supabase functions contain an incorrect authorization flaw in the API-key bundle promotion path. The PUT /bundle endpoint, available to "all" and "write" API keys, dispatches to setChannel, which authorizes with checkPermission(c, 'channel.promote_bundle', { appId: body.app_id }) and omits the request's channel_id. Because the omitted scope field is passed to rbac_check_permission_direct as SQL NULL, and channel-scope override evaluation is gated on p_channel_id IS NOT NULL, per-channel allow/deny overrides are never evaluated. A principal holding app-level channel.promote_bundle (granted by default to the app_developer and app_uploader roles) can therefore promote a bundle to a channel for which an explicit per-channel deny override exists, updating public.channels.version for the supplied channel_id; the target channel is only validated after authorization. The issue is confirmed on main at commit de66fa51e7ff2f50283cc1455c3d80ab3eb0ae43 and likely earlier versions; no patched version is known at the time of publication.

Capgo (capgo.app) exposes a native build TUS upload proxy (supabase/functions/_backend/public/build/upload.ts) that authorizes a caller against a single build job identified by the supplied builder_job_id and validates only that job's stored upload_path, but then forwards the user-controlled TUS resource suffix taken from /build/upload/:jobId/* to the builder service while injecting Capgo's privileged builder API key. Because the forwarded suffix is never bound to the authorized job's upload_path or upload_session_key, a caller holding a valid 'all' or 'write' Capgo API key with app.build_native permission for one application can use its authorized proxy path for job A to write to the TUS upload resource of another job B, provided that resource suffix is known or exposed, corrupting that build's artifacts. All versions are affected; no patch was available at the time of advisory publication.

Capgo.app before 12.264.5 does not enforce upload expiry or build lifecycle state in the /build/upload/:jobId TUS proxy endpoint. When a native build request is created, an upload_expires_at timestamp (one hour) and a 'pending' status are stored in build_requests, but the upload proxy loads only app_id, owner_org, builder_job_id, and upload_path and checks only the app.build_native permission before forwarding POST, PATCH, and HEAD requests to the internal builder. As a result, an authenticated caller holding app.build_native permission for the app can continue writing to the build upload session after the stored expiry has passed or after the build has moved beyond the upload phase, unless the separate builder service independently rejects the request. The issue is fixed in 12.264.5.

Capgo (capgo.app) exposes the legacy membership table public.org_users directly through Supabase PostgREST. The table's row-level security policies "Allow org admin to insert" and "Allow org admin to update" only verify that the caller has admin rights in the target organization (public.check_min_rights('admin', ...)); they do not require a pending invitation in tmp_users, acceptance of an invite token via /private/accept_invitation, any action by the target user, or the membership/role-consistency and anti-escalation checks enforced by the RBAC role-binding path. As a result, an authenticated user who is an admin of an organization can INSERT or UPDATE org_users rows directly to add any existing public.users account as an active member of that organization with user_right="admin", bypassing the invitation and role-assignment workflow entirely. In testing, an account with no prior access to the organization or its apps could, after such a direct insert, read the organization and app and pass check_min_rights. All versions are affected and no patch was available at the time of publication.

capgo.app through 12.129.0 fails to verify deletion status when serving cached bundle artifacts from the public file read endpoint. Unauthenticated attackers can download deleted bundles using cached URLs and trigger restoration of deleted objects into R2 storage on cache hits.

Capgo (capgo.app) contains an incomplete access-control/content-lock enforcement issue affecting all versions; no patch is available at the time of publication. The `enforce_encrypted_bundle_trigger` / `check_encrypted_bundle_on_insert` content lock in supabase/schemas/prod.sql exempts `app_versions` rows whose previous `storage_provider` is `r2-direct`, so delivery-critical columns (name, app_id, session_key, key_id, storage_provider, r2_path, external_url, checksum, manifest, native_packages) remain mutable while a row is in that state. Separately, the channel/version join used by the `/updates` endpoint (supabase/functions/_backend/utils/pg.ts) filters only deleted versions and does not exclude `r2-direct` versions, so such rows remain OTA-selectable. An attacker holding a Capgo API key with bundle write/upload permission can create or keep a bundle version in the `r2-direct` state, assign it to a channel, and then modify these fields via direct PostgREST requests, causing altered bundle metadata (for example checksum and session_key) to be served to devices through `/updates`. This bypasses the post-upload content immutability hardening added for a previous app_versions metadata-tampering issue.

Capgo CLI (npm package @capgo/cli) through 7.98.2 is affected by an over-permissioned service account in its Android onboarding flow. When onboarding via Google OAuth, the CLI invites the generated Google Play service account with the account-wide Play Console permission CAN_MANAGE_DRAFT_APPS_GLOBAL (passed as developerAccountPermissions in the Android Publisher API User create request), even though the user-facing flow states the service account is invited into a single confirmed app with release-only permissions. As a result, anyone who obtains the generated service account key (PLAY_CONFIG_JSON) can create, edit, and delete draft apps across the entire Google Play developer account rather than being limited to the selected package. No patched version was available at the time of publication.

Capgo (capgo.app) blocks direct user inserts into the public.manifest table with a RESTRICTIVE row-level security policy, but that restriction can be bypassed indirectly. A principal holding an app-scoped upload/write/all API key (upload+ rights) or an authenticated user with write+ rights on an app can update public.app_versions.manifest on a version whose storage_provider is 'r2-direct', which is not covered by the bundle content-lock check. The on_version_update async worker trusts record.manifest and, using the service-role Supabase client, inserts the attacker-controlled file_name, file_hash, and s3_path into public.manifest before clearing app_versions.manifest. When a channel points to the crafted version, the /updates endpoint returns the service-role-created manifest entry as a client-facing download_url, enabling OTA manifest poisoning through a trusted async worker path. All versions are affected; no patch was available at the time of publication.

Capgo (capgo.app) is affected by an authorization flaw in the app icon update path. The PUT /app/:id endpoint accepts a user-controlled `icon` value, normalizes it, and stores it in public.apps.icon_url without verifying that the image path belongs to the target app's own image namespace (e.g. org/{owner_org}/{app_id}/...). Updating apps.icon_url fires the on_app_update trigger, whose worker reads record.icon_url and calls cleanStoredImageMetadata(), which runs with service-role credentials (supabaseAdmin()) and downloads and re-uploads the referenced storage object with upsert: true. As a result, an authenticated holder of an app-limited write API key can cause the privileged worker to rewrite an out-of-scope private image object (for example an organization logo) that the key cannot read or write directly under Supabase Storage RLS. All versions are affected; no patched version was available at the time of the advisory.

Cap-go capgo.app fails to validate that principals in channel_permission_overrides belong to the organization, allowing authenticated app/org admins to grant channel permissions to non-member users. Attackers with admin privileges can insert override rows with arbitrary external user UUIDs to grant channel-scoped permissions such as channel.promote_bundle to users outside the organization.

capgo.app is an over-the-air update platform for Capacitor apps. In all versions prior to a fix, the row-level security UPDATE policy on the public.orgs table permits an organization admin (a user holding org.update_settings) to update the entire row, including the internal billing pointer column customer_id. The official organization update endpoint (supabase/functions/_backend/public/organization/put.ts) allowlists only a small set of editable settings fields and excludes customer_id, and the private Stripe billing route separately requires the org.update_billing permission. By sending an update directly to Supabase PostgREST, an authenticated org admin without org.update_billing can null or corrupt the organization's Stripe customer pointer, causing plan and billing checks that trust orgs.customer_id to fail and moving the organization from a valid paid plan state to unpaid/no-plan behavior. At the time of the advisory no patched version was available.

Cap-go capgo.app before 12.267.1 fails to validate target API key privilege during rotation, allowing an apikey_manager to rotate a higher-privileged org_super_admin sibling key and recover its plaintext credential. Attackers with apikey_manager role can enumerate same-owner API keys, rotate a stronger sibling through the PUT endpoint, and obtain the replacement plaintext secret to authenticate as the higher-privileged principal.

Capgo before 12.244.1 contains a cross-tenant integrity vulnerability in the metadata-cleaning worker that trusts image object keys from mutable database rows without validating ownership. An authenticated attacker can place a victim tenant's image key in a row they control, causing the service-role worker to download and re-upload that object with sanitized metadata. Attackers can silently modify metadata in cross-tenant image objects by supplying known victim keys during authorized row updates, bypassing storage access controls through the confused-deputy metadata worker.

capgo.app is an over-the-air (OTA) update platform for Capacitor apps. In all versions up to and including the current release (no patch available at time of publication), the `transfer_app()` database function transfers an app, its channels, versions and related records to a destination organization without deleting or revalidating existing rows in `channel_permission_overrides`. As a result, a user who legitimately held a channel permission override while a member of the source organization retains that override after the transfer, even though they have no membership and no RBAC binding in the destination organization. Using their own authenticated JWT against the PostgREST API, such a former member can modify the destination-owned channel to point at a different bundle, causing the /updates endpoint to serve an attacker-selected application version to devices. The previously proposed fix for GHSA-626c-p6fq-3whq (PR #3093), which validates organization membership when an override is created or updated, does not remove overrides that became stale as a result of an app transfer.

Capgo (capgo.app) through version 12.261.0 contains an incomplete access-control fix for the public.sso_providers table. Migration 20260826100000_sso_providers_block_direct_active_insert.sql installs a BEFORE UPDATE guard (enforce_sso_provider_client_update_guard()) that freezes only the dns_verified_at, domain, status and enforce_sso columns; provider_id (as well as metadata_url and attribute_mapping) is left writable. Because the table is granted ALL to the anon and authenticated roles with no column-level restriction, and PostgreSQL row-level security policies such as allow_org_admins_update_sso_providers constrain only which row may be updated and not which columns, a user holding the org_admin tier permission org.update_settings can PATCH provider_id over PostgREST to an identity provider under their control. Since provider_id is the trust anchor binding an email domain to an authorized IdP, the attacker can then authenticate through their own IdP while asserting the org owner's email; the server-side provider match succeeds and the merge routine attaches the attacker's SSO identity to the existing owner account, nulls its password, and deletes its other identities and sessions. This results in vertical privilege escalation from org_admin to org owner/super_admin, account takeover, and lockout of the legitimate owner. Exploitation requires that the target organization has an active SSO provider configured and that the attacker already holds org_admin in that organization. No patched version is available.

Capgo (capgo.app backend, versions ≤ 12.261.0) improperly restricts which roles the apikey_manager organization role may bind to newly created API keys. When an authenticated user holding only apikey_manager (permissions org.manage_apikeys and org.read) calls POST /apikey with a JWT session, the only checks applied are the org.manage_apikeys permission, a fixed deny-list of assignable role names (APIKEY_MANAGER_DENIED_ASSIGNABLE_ROLES in public/apikey/scope.ts), and a priority-rank comparison in createRoleBindingForPrincipal (private/role_bindings.ts). No check verifies that the caller actually holds the permissions conferred by the role being assigned. Because the deny-list omits the deploy roles app_developer, app_uploader, channel_developer and channel_uploader, and apikey_manager is seeded with priority_rank 78 — higher than those roles' ranks (68, 66, 58, 57) — the rank check also passes. As a result, an apikey_manager who cannot upload bundles or promote channels can mint an API key bound to a deploy role and use it to push arbitrary OTA JavaScript updates to all end users of the organization's apps. As of the advisory publication no patched version was available.

Flowise through 3.1.4 exposes GET /api/v1/upsert-history/:id and PATCH /api/v1/upsert-history without route-level permission checks, and the backing service performs no workspace or ownership validation. getAllUpsertHistory() returns UpsertHistory rows selected solely by an attacker-supplied chatflowid, and patchDeleteUpsertHistory() deletes rows by an attacker-supplied array of record UUIDs. As a result, any authenticated low-privilege user or valid API key can read or delete document-store upsert history belonging to other users and other workspaces whenever the target chatflowId (which is exposed publicly in /chatbot/<chatflowId> share links) or row ids are known. The retrievable flowData and result fields contain embedding, record-manager and vector-store node configuration, including per-node paramValues. No patched version is available.

Flowise (npm packages `flowise` and `flowise-components`) through 3.1.4 looks up credentials by ID without filtering on the requesting user's workspace (findOneBy({ id: credentialId }) with no workspaceId condition) in several code paths: getAllOpenaiAssistants/getSingleOpenaiAssistant (GET /api/v1/openai-assistants and /api/v1/openai-assistants/:id), uploadFilesToAssistant (POST /api/v1/openai-assistants-file/upload/), deleteAssistant (DELETE /api/v1/assistants/:id, reachable by first importing a poisoned assistant row via POST /api/v1/export-import/import), and the shared helper used by getVoices (GET /api/v1/text-to-speech/voices). An authenticated user of one workspace can supply a credential UUID belonging to another workspace, causing the server to decrypt and use that workspace's OpenAI or ElevenLabs API key on the attacker's behalf. No patched version was available at the time of publication.

Flowise through 3.1.4 does not enforce authorization on the BullMQ admin dashboard. When the server runs in queue mode with the dashboard enabled and not in cloud mode (MODE=queue, ENABLE_BULLMQ_DASHBOARD=true, and !isCloud()), the /admin/queues mount is protected only by the verifyTokenForBullMQDashboard middleware, which validates the JWT but performs no role, permission, or workspace/organization scoping check; the mount also lies outside /api/v1/* so the global API gate does not apply. As a result, any authenticated user — including the lowest-privileged member of any tenant — can reach the full Bull-Board UI and view all queues and job payloads across the entire instance, including chat inputs and overrideConfig (which may carry credentials and prompts), chatflow.flowData graph definitions with custom function source code, credential IDs and system prompts, chatIds, files, and the originating orgId/workspaceId. The dashboard's write actions (retry, remove, promote, clean) are likewise usable across tenants. No patched version is available as of the advisory.

Flowise through 3.1.4 resolves SSO and local-password users solely by email without storing provider or subject identifier bindings, allowing attackers to authenticate as any existing user by claiming their email at any configured SSO provider. Attackers can gain complete account access including chatflows, credentials, and API keys by authenticating through a different SSO provider or local password than the victim's original registration method.

Flowise through 3.1.4 (Enterprise/platform mode with SSO enabled) contains an authentication bypass in the SSO login path. When an SSO callback arrives with an email matching a user whose status is INVITED, verifyAndLogin (SSOBase.ts:80-94) copies the user record from the database — including the server-stored single-use invitation tempToken — into the data passed to AccountService.register(). The register handler's token lookup, email match, and expiry checks therefore pass trivially against the server's own token instead of a caller-supplied one, and the account and its organization membership are flipped to ACTIVE. As a result, anyone able to authenticate at any configured SSO provider using a pending invitee's email address as the email claim can take over that invitation and obtain the invited user's access to the organization without ever possessing the emailed invitation token, for as long as the invitation is valid (24 hours by default). At the time of the advisory no patched version was available.

Flowise through 3.1.4 contains missing route-level RBAC checks on chat message endpoints that allow low-privileged API keys to read and delete chat history. Attackers with valid but low-privileged API keys can access GET and DELETE chat message routes without required flow permissions to read chat histories, prompts, model responses, and delete messages.

ClawHub (openclaw/clawhub) contains an incorrect authorization vulnerability in the ClawHub application/backend: an organization-owned skill retains the ownerUserId of its original publisher, and transfer and lifecycle authorization checks trust that historical user before requiring current organization privileges. An authenticated user who originally published an organization skill can therefore transfer, delete, or restore that skill — taking control of its trusted name and history — even after their organization privileges have been revoked or downgraded. The issue was confirmed at revision cbfee7343ddc867316dd9b3de6fa8856730f9f41; the complete historical affected range was not established. It is fixed by PR #3680, included in revision 8c2de6c506bb4efabe3f0c2ffb8370b9e23d4650, which was deployed to clawhub.ai on 2026-09-11; self-hosted deployments should update to that revision or a later descendant. The npm CLI and OpenClaw runtime are separate products and are not affected.

ClawHub (openclaw/clawhub) application/backend contains a flaw in the skill report moderation flow: four distinct ordinary authenticated accounts can report a visible skill and trigger automatic hiding (moderationStatus: hidden) of that skill from the catalog without any moderator decision. Because the reporter quota counts only reports filed against visible targets, the same accounts can repeat the process against additional skills; official skills are not exempt. The issue was confirmed at revision cbfee7343ddc867316dd9b3de6fa8856730f9f41; the complete historical affected range was not established. The fix (PR #3681) is included in revision 8c2de6c506bb4efabe3f0c2ffb8370b9e23d4650; self-hosted deployments should update to that revision or a later descendant. The npm CLI and OpenClaw runtime are separate products and are not affected.

ClawHub (openclaw/clawhub application/backend) contains a missing authorization check in the changelog preview feature. A signed-in caller can invoke the public skills:generateChangelogPreview action for a skill they are not authorized to access; the previous version is read without the file-read authorization enforced on normal content access, and up to 8,000 characters of quarantined content may be submitted to the AI provider and reflected in the preview returned to the caller, disclosing restricted skill content. The issue was confirmed at revision cbfee7343ddc867316dd9b3de6fa8856730f9f41; the complete historical affected range was not established. It is fixed by PR #3682, included in revision 8c2de6c506bb4efabe3f0c2ffb8370b9e23d4650, which was deployed to clawhub.ai on 2026-09-11; self-hosted deployments should update to that revision or a later descendant. The npm CLI and OpenClaw runtime are separate products and are not affected.

ClawHub (openclaw/clawhub) application/backend contains a server-side request forgery vulnerability in the public profile preview's image fetching. The preview accepts a user-supplied image URL and checks the textual hostname against private-address patterns, but does not validate or pin the resolved network destination, so a public-looking hostname can resolve to an internal address or change resolution between validation and connection (DNS rebinding). A maintainer-run local harness demonstrated an outbound connection to an owner-controlled loopback listener; access to production internal services, credential disclosure, and code execution were not demonstrated. The issue was confirmed at revision cbfee7343ddc867316dd9b3de6fa8856730f9f41; the complete historical affected range was not established. Fixed by PR #3683, included in revision 8c2de6c506bb4efabe3f0c2ffb8370b9e23d4650, which was deployed to clawhub.ai on 2026-09-11; self-hosted deployments should update to that revision or a later descendant. The npm CLI and OpenClaw runtime are separate products and are not affected.

ClawHub (the openclaw/clawhub application/backend) does not bind anonymous HTTP API requests to a trusted caller identity, so all direct anonymous API requests share a single default quota allowance. A remote, unauthenticated caller can drain that shared allowance and thereby deny or degrade API access for unrelated visitors. In addition, when the TRUST_FORWARDED_IPS option is enabled without an authenticated edge/proxy, clients can supply arbitrary forwarded IP headers to select quota identities of their choosing and evade rate limiting. The issue was confirmed at revision cbfee7343ddc867316dd9b3de6fa8856730f9f41; the complete historical affected range was not established. It is fixed in revision 8c2de6c506bb4efabe3f0c2ffb8370b9e23d4650 (PR #3684), where direct anonymous calls are redirected to the public API origin without consuming quota and forged identity assertions return HTTP 401. The npm CLI and OpenClaw runtime are separate products and are not affected.