Open WebUI: Read-Only Users Can Modify Collaborative Documents via Socket.IO
Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. Prior to 0.9.0, the ydoc:document:update Socket.IO event handler checks whether the sender is a member of the document's Socket.IO room (line 678) but does not verify that the sender has write permission. Users with read-only access join the document room via ydoc:document:join, which only requires read permission (line 520). Once in the room, the user can emit ydoc:document:update events that modify the in-memory Yjs document state and are broadcast to all other collaborators in real time. This vulnerability is fixed in 0.9.0.
Open WebUI: Stored XSS in Pending User Overlay via Incorrect DOMPurify Application Order
Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. Prior to 0.9.0, the AccountPending.svelte component renders the admin-configured "Pending User Overlay Content" using marked.parse() inside {@html} with an incorrect DOMPurify application order. An admin can inject arbitrary JavaScript into the Pending User Overlay Content that executes in the browser context of any pending user who views the overlay page. This vulnerability is fixed in 0.9.0.
Open WebUI: Full SSRF Vulnerability in the RAG Web Search Feature
Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. Prior to 0.9.0, validate_url() in backend/open_webui/retrieval/web/utils.py calls validators.ipv6(ip, private=True), but the validators library does NOT implement the private keyword for IPv6 — the call raises a ValidationError (which is falsy in a boolean context), so every IPv6 address passes the filter. In addition, IPv4-mapped IPv6 (::ffff:10.0.0.1) bypasses the IPv4 check entirely, and several reserved IPv4 ranges (0.0.0.0/8, 100.64.0.0/10, 192.0.0.0/24, etc.) are not blocked. This vulnerability is fixed in 0.9.0.
Open WebUI: API key endpoint restrictions bypassed via `x-api-key` header — full message processing on restricted endpoints
Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. Prior to 0.9.0, Open WebUI allows admins to restrict which API endpoints an API key can access. When an API key is restricted from /api/v1/messages, requests using the Authorization: Bearer sk-... header are correctly blocked with 403. However, the same key sent via the x-api-key header bypasses the restriction entirely — the request is authenticated, the model is invoked, and a full response is returned. This vulnerability is fixed in 0.9.0.
Open WebUI: Broken Access Control for Completions API
Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. Prior to 0.9.0, a user just needs to use the API endpoint: /api/chat/completions with their own API key (generated in OWUI) and the Chat ID of another user to continue the conversation of the other user. This vulnerability is fixed in 0.9.0.
Open WebUI: Low-privilege authenticated users can enumerate and stop global background tasks, causing system-wide chat disruption
Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. Prior to 0.9.0, any authenticated user with low privileges can enumerate active background tasks across the system and stop tasks belonging to other users via the GET /api/tasks and POST /api/tasks/stop/{task_id} methods. This allows a casual user to disrupt system-wide chat usage by continuously canceling other users' active tasks. This is a real authorization vulnerability affecting integrity and usability in multi-user deployments. This vulnerability is fixed in 0.9.0.
Open WebUI: shared-chat branch ignores access_type, allowing unauthorized file deletion
Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. Prior to 0.9.0, any authenticated user can permanently delete files owned by other users via DELETE /api/v1/files/{id} when the target file is referenced in any shared chat. The has_access_to_file() authorization gate unconditionally grants access through its shared-chat branch. It checks neither the requesting user's identity nor the type of operation being performed. File UUIDs (which would otherwise be impractical to guess) are disclosed to any user with read access to a knowledge base via GET /api/v1/knowledge/{id}/files. This vulnerability
Open WebUI: LDAP and OAuth First-User Race Condition Allows Multiple Admin Accounts
Open WebUI is a self-hosted artificial intelligence platform designed to operate entirely offline. Prior to 0.9.0, the LDAP and OAuth authentication flows use a TOCTOU (Time-of-Check-Time-of-Use) pattern for first-user admin role assignment. The regular signup handler (signup_handler in auths.py, line 663) was explicitly patched to prevent this race with the comment "Insert with default role first to avoid TOCTOU race", but the LDAP and OAuth code paths were never updated with the same fix. This vulnerability is fixed in 0.9.0.
Microsoft APM: Windows absolute-path tar member overwrite during legacy-bundle probing in `apm install`
Microsoft APM is an open-source, community-driven dependency manager for AI agents. Prior to 0.13.0, Microsoft APM contains a Windows-specific archive extraction boundary failure in the legacy-bundle probe used by apm install <bundle> on supported Python 3.10 and 3.11 runtimes. When apm install is given a local .tar.gz that is not recognized as a plugin-format bundle, APM probes whether it is a legacy --format apm bundle. On Python versions earlier than 3.12, that probe extracts untrusted tar members with raw tar.extractall() without rejecting Windows absolute member names such as D:/.... This vulnerability is fixed in 0.13.0.
Microsoft APM: Symlinks under `.apm/prompts/` and `.apm/agents/` are dereferenced during `apm install`, copying host-local file contents into the project tree
Microsoft APM is an open-source, community-driven dependency manager for AI agents. From 0.5.4 to 0.12.4, two primitive integrators in apm-cli enumerate package files with bare Path.glob() / Path.rglob() calls and read each match with Path.read_text(), transparently following symbolic links. A symlink committed inside a remote APM dependency under .apm/prompts/<x>.prompt.md or .apm/agents/<x>.agent.md is preserved verbatim into apm_modules/ on clone and then dereferenced during integration, with the resolved content written as a regular file into the project's deploy directories. The package content_hash, the pre-deploy SecurityGate scan, and
Microsoft APM: plugin.json component paths escape plugin root and copy arbitrary host files during install
Microsoft APM is an open-source, community-driven dependency manager for AI agents. Prior to 0.8.12, Microsoft APM normalizes marketplace plugins by copying plugin components referenced in plugin.json into .apm/. The manifest fields agents, skills, commands, and hooks are attacker-controlled, but the implementation does not enforce that those paths remain inside the plugin directory. A malicious plugin can therefore use absolute paths or ../ traversal paths to copy arbitrary readable host files or directories from the installer's machine during apm install. This vulnerability is fixed in 0.8.12.
A vulnerability in mlflow/mlflow versions 3.9.0 and earlier allows unauthenticated access to certain FastAPI routes when the server is started with authentication enabled (`--app-name basic-auth`) and served via uvicorn (ASGI). The FastAPI permission middleware only enforces authentication on `/gateway/` routes, leaving other routes such as the Job API (`/ajax-api/3.0/jobs/*`) and the OpenTelemetry trace ingestion API (`/v1/traces`) unprotected. This allows unauthenticated remote attackers to submit jobs, read job results, cancel running jobs, and inject arbitrary trace data into experiments. The issue arises from an architectural mismatch be
Home Assistant MCP Server: YAML config backups written under www/ are served unauthenticated at /local/
### Summary
When `ENABLE_YAML_CONFIG_EDITING=true`, every `ha_config_set_yaml` call backs up the pre-edit file to `<config>/www/yaml_backups/`, which Home Assistant serves at `/local/` with **no authentication**. Anyone who can reach the HA web interface can download the most recent pre-edit `configuration.yaml` (or other YAML file) — typically containing plaintext MQTT passwords, REST credentials, webhook IDs, geofence coordinates, and `shell_command` definitions — with zero credentials.
### Details
The backup feature is good — `do_backup` defaults to `True` and protects users from a bad edit. The issue is the location:
- `custom_compone
python-utcp: Command Injection via Unsanitized Argument Substitution in CLI Communication Protocol
python-utcp is the python implementation of UTCP. Prior to 1.1.3, the _substitute_utcp_args method in cli_communication_protocol.py inserts user-controlled tool_args values directly into shell command strings without any sanitization or escaping. These commands are then executed via /bin/bash -c (Unix) or powershell.exe -Command (Windows), allowing an attacker to inject arbitrary shell commands. This vulnerability is fixed in 1.1.3.
python-utcp: Full Process Environment Exposed to CLI Subprocess - Secrets Leakage via Command Injection
python-utcp is the python implementation of UTCP. Prior to 1.1.3, _prepare_environment() in cli_communication_protocol.py passes a full copy of os.environ to every CLI subprocess. When combined with CVE-2026-45369, an attacker can exfiltrate all process-level secrets in a single tool call. This vulnerability is fixed in 1.1.3.
python-utcp: SSRF via attacker-controlled OpenAPI servers[0].url in HTTP communication protocol
python-utcp is the python implementation of UTCP. Prior to 1.1.3, the utcp-http plugin is vulnerable to a blind Server-Side Request Forgery (SSRF) caused by a trust-boundary inconsistency between manual discovery and tool invocation. register_manual() validates the discovery URL against an HTTPS / loopback allowlist, but call_tool() and call_tool_streaming() reuse the resolved tool_call_template.url directly without revalidating, and the OpenAPI converter blindly trusts whatever servers[0].url an attacker-hosted spec declares. An attacker who hosts a malicious OpenAPI spec on a legitimate HTTPS endpoint can declare e.g. servers: [{ url: "http
Missing integrity verification in Triton inference handler in Amazon SageMaker Python SDK
Missing integrity verification in the Triton inference handler in Amazon SageMaker Python SDK v2 before v2.257.2 and v3 before v3.8.0 might allow a remote authenticated actor to achieve code execution in inference containers via replacement of model artifacts in S3 with a specially crafted pickle payload that is deserialized without verification. This issue requires a remote authenticated actor with S3 write access to the model artifact path.
To remediate this issue, we recommend upgrading to Amazon SageMaker Python SDK v2.257.2 or v3.8.0 and rebuild any Triton models previously created with ModelBuilder using the updated SDK.
Cleartext storage of HMAC signing key in Amazon SageMaker Python SDK ModelBuilder/Serve path
Cleartext storage of sensitive information in the ModelBuilder/Serve component in Amazon SageMaker Python SDK before v2.257.2 and v3 before v3.8.0 might allow a remote authenticated actor to extract the HMAC signing key from SageMaker API responses and forge valid integrity signatures for specially crafted model artifacts, achieving code execution in inference containers. This issue requires a remote authenticated actor with permissions to call SageMaker describe APIs and S3 write access to the model artifact path.
To remediate this issue, we recommend upgrading to Amazon SageMaker Python SDK v2.257.2 or v3.8.0 and rebuild any models previ
Docling-Graph: SSRF via Missing Internal IP Validation in URLInputHandler
Docling-Graph turns documents into validated Pydantic objects, then builds a directed knowledge graph with explicit semantic relationships. Prior to 1.5.1, the URLInputHandler class in docling_graph/core/input/handlers.py makes HTTP requests to user-supplied URLs without validating whether the target resolves to a private, loopback, or link-local IP address. The URLValidator only checks for a valid scheme and non-empty netloc, performing no IP-level validation. Additionally, requests.head() was called with allow_redirects=True, allowing an attacker to redirect requests to internal endpoints via an intermediary URL. An attacker who can control
Diffusers is the a library for pretrained diffusion models. Prior to 0.38.0, diffusers 0.37.0 allows remote code execution without the trust_remote_code=True safeguard when loading pipelines from Hugging Face Hub repositories. The _resolve_custom_pipeline_and_cls function in pipeline_loading_utils.py performs string interpolation on the custom_pipeline parameter using f"{custom_pipeline}.py". When custom_pipeline is not supplied by the user, it defaults to None, which Python interpolates as the literal string "None.py". If an attacker publishes a Hub repository containing a file named None.py with a class that subclasses DiffusionPipeline, t