Meta Ads MCP: Unauthenticated HTTP MCP Tool Execution Leaks Operator Meta Access Token
Meta Ads MCP is a Model Context Protocol (MCP) server that lets AI assistants run Meta Ads. Prior to version 1.0.109, `AuthInjectionMiddleware.dispatch()` at `http_auth_integration.py:272` unconditionally forwards unauthenticated Streamable HTTP requests to downstream MCP tool handlers without issuing a `401` response, allowing any network-reachable caller to invoke MCP tools without authentication. When no per-request credential is present, tool handlers fall back to the `META_ACCESS_TOKEN` environment variable, and when the downstream Meta Graph API call fails, `api.py:263–269` serialises the raw `httpx` request URL—including the operator's
pypdf: Possible large memory usage for large /ToUnicode streams
pypdf is a free and open-source pure-python PDF library. Prior to 6.15.0, a crafted PDF can cause large memory consumption when pypdf/_cmap.py function parse_bfrange parses unusually large source-code or destination-string tokens in a font /ToUnicode CMap during text extraction. This issue is fixed in 6.15.0.
pypdf: Possible long runtimes/large memory usage for large CID font width ranges
pypdf is a free and open-source pure-python PDF library. Prior to 6.15.0, a crafted PDF can cause long runtimes and large memory consumption when pypdf/_font.py function Font._collect_cid_character_widths expands unusually large CID font /W width ranges or excessive width entries during text extraction. This issue is fixed in 6.15.0.
A vulnerability in `nltk.downloader` in nltk/nltk versions <= 3.9.4 allows for cross-package resource and model poisoning. The downloader extracts package archives into shared namespaces such as `corpora/` and `taggers/` instead of package-isolated roots, and validates package integrity only after the archive has been written and extracted. This design flaw enables one package to overwrite another package's trusted resources within the same namespace, making the changes immediately active through ordinary NLTK APIs. This issue persists across fresh interpreter restarts and can affect downstream workflows, including machine learning pipelines
h2: Duplicate Host header could facilitate request smuggling
h2 is a pure-Python implementation of a HTTP/2 protocol stack. Versions up to and including 4.4.0 accept request header blocks containing more than one Host header, and forward every Host header to the consuming application. Where the consumer downgrades HTTP/2 to HTTP/1.1, the resulting request carries two Host header lines, providing a request smuggling primitive. This issue is fixed in version 4.4.1.
PyMdown Extensions: Path traversal in the b64 extension lets <img src> read files outside base_path
PyMdown Extensions is a set of extensions for the Python-Markdown markdown project. In versions up to and including 10.21.3, the b64 extension is vulnerable to a path traversal that discloses arbitrary files: it inlines images referenced by <img src="..."> by joining the src onto the configured base_path with os.path.normpath and opening the result directly, without verifying that the resolved path stays inside base_path. As a result, an src containing ../ sequences or an absolute path reads a file outside base_path as long as it has an allowed image extension (.png, .jpg, .jpeg, .gif, .svg), and the file's contents are then base64-encoded in
pymdown-extensions: Exponential-backtracking ReDoS in caret, tilde, betterem, and magiclink inline processors
pymdown-extensions is a collection of extensions for the Python Markdown library. In versions up to and including 11.0, four inline processors (caret, tilde, betterem, and magiclink) use regular expressions whose content groups can partition a run of delimiter characters in exponentially many ways, causing catastrophic backtracking. As a result, a single untrusted Markdown line under 50 bytes rendered with markdown.markdown() in each extension's default configuration drives the rendering thread into unbounded CPU usage that grows exponentially with input length, enabling an unauthenticated remote attacker who can submit Markdown to cause deni
LangGraph: Namespace prefix matching crosses segment boundaries in Postgres and SQLite stores
LangGraph Checkpoint Postgres and SQLite Checkpoint are the Postgres and SQLite implementations of LangGraph's checkpoint saver. Prior to 3.1.1, the langgraph-checkpoint-postgres and langgraph-checkpoint-sqlite packages persisted hierarchical namespaces as a dot joined string and scoped reads by matching that string as a simple prefix pattern, so a read scoped to one namespace could also match a sibling namespace whose flattened form shares the same leading characters, or a namespace label containing unescaped pattern metacharacters, allowing an authenticated caller to retrieve stored items belonging to another tenant or user through an ordin
aiosend: Deserialization of request body before signature verification (Pre-auth DoS) in webhook handler
aiosend is a synchronous and asynchronous Crypto Pay API client. Pror to version 3.0.7, `WebhookHandler.feed_update()` deserializes the entire request body before verifying the HMAC signature. This allows an unauthenticated attacker to force expensive parsing of arbitrary JSON payloads that will ultimately be rejected, leading to unnecessary CPU and memory consumption. Version 3.0.7 fixes the issue. Some workarounds are available. Restrict request body size at the reverse proxy or web framework, rate-limit webhook endpoints, and/or reject oversized requests before JSON parsing.
Apache Polaris did not consistently validate storage locations supplied during table and view registration.
An authenticated principal with permission to register a table or view could, depending on the affected release and registration path, cause Polaris to use the catalog's storage credentials to read a caller-selected Iceberg metadata file before verifying that the file was within the catalog's allowed storage locations.
If the catalog's underlying credentials could read an object outside that boundary, this could disclose limited information from the object.
Polaris could also accept registration metadata located within an allowed lo
PraisonAI: SSRF in web_crawl tool via redirect-following and DNS rebinding (validate-then-fetch gap)
PraisonAI is a multi-agent teams system. In versions prior to 1.6.58, the web_crawl tool performs its SSRF check only on the initially supplied URL, allowing the protection to be bypassed so the tool connects to attacker-chosen internal destinations. The check resolves the hostname once with socket.gethostbyname and rejects private/loopback/link-local results, but then passes the URL to a fetcher using httpx.Client(follow_redirects=True) (or urllib.request.urlopen when httpx is absent, which also follows redirects) that re-resolves the hostname at connect time with no further validation. This validate-here/fetch-there gap is exploitable throu
PraisonAI has a`web_crawl` SSRF protection bypass via unchecked redirect targets
PraisonAI is a multi-agent teams system. In versions 1.5.128 through 1.6.57, the praisonaiagents.tools.web_crawl_tools.web_crawl() function is vulnerable to server-side request forgery. While it validates the initially supplied URL and blocks direct loopback and private destinations, its default httpx fallback uses httpx.Client(follow_redirects=True) and does not revalidate intermediate or final redirect targets. An attacker who can influence a URL passed to web_crawl(), directly or through an agent or tool workflow, can supply an attacker-controlled public URL that passes the initial host check and then redirects to loopback, private-network
PraisonAI workflow include bypasses tools.py autoload opt-in and executes included recipe code
PraisonAI is a multi-agent teams system. In versions 3.9.26 through 4.6.57 of praiseonai and 0.12.12 through 1.6.57 of praiseonaiagents, the workflow "include" feature is vulnerable to code execution. Workflow._execute_include() implicitly imports and runs an included recipe's tools.py via a raw importlib.util.spec_from_file_location() and spec.loader.exec_module() call, without honoring the PRAISONAI_ALLOW_TEMPLATE_TOOLS/PRAISONAI_ALLOW_LOCAL_TOOLS autoload opt-in gates or routing through the centralized safe loader that protects the other tools.py autoload paths. As a result, a workflow that includes an attacker-controlled local recipe dire
MLflow's AI Gateway accepts an auth_config.api_base value when creating a gateway secret (mlflow/server/handlers.py, _create_gateway_secret) with no validation of scheme, host, or IP range; the value is stored verbatim. The gateway proxy endpoint (mlflow/server/gateway_api.py, raw_proxy) subsequently issues an HTTP request to that stored api_base plus a caller-supplied path and returns the full response body.
Open WebUI: A folder write-collaborator can permanently delete the owner's chats by deleting a shared subfolder
Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. From 0.10.0 until 0.11.0, the DELETE /api/v1/folders/{id} handler in backend/open_webui/routers/folders.py allowed a user granted write access to a shared chat folder to permanently delete chats and messages belonging to the folder owner. The cascade following the authorization check is bound to the folder owner's id, but the subfolder check accepted any inherited write grant instead of requiring ownership or administrator status. A collaborator can destroy the owner's subtree or force-move chats out of it when delete_contents=false. This issue is fixed in 0
Open WebUI: Any authenticated user can stall a worker via a knowledge-search pattern that backtracks catastrophically
Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. From 0.9.6 until 0.11.0, the built-in knowledge search path in backend/open_webui/tools/knowledge_fs.py and backend/open_webui/tools/builtin.py let a chat participant choose a pattern used to grep knowledge files. Patterns containing regex metacharacters were compiled with Python's backtracking re engine and run against every line of every reachable file with no time limit, so a crafted pattern such as (x|x)*y and one matching uploaded file line can pin one CPU core and block the event loop. This causes availability impact for every other user of the affecte
Open WebUI: Stored XSS via unescaped KaTeX render-error fallback in rendered messages
Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. From 0.10.0 until 0.11.0, src/lib/components/chat/Messages/Markdown/KatexRenderer.svelte could store and render a chat message whose math block makes KaTeX fail with a stack overflow instead of a parse error. The catch branch fell back to inserting the original math source into the page as HTML through {@html} rather than as text, so script in the message runs in the browser of whoever views it, including shared chats and channels. The viewer's session token in localStorage can be stolen, and an administrator viewer can have their account taken over. This is
Open WebUI: Tool source code disclosed to read-only users via the tool list and get endpoints
Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. In 0.10.2 and earlier, the GET /api/v1/tools/, GET /api/v1/tools/list, and GET /api/v1/tools/id/{id} endpoints in backend/open_webui/routers/tools.py returned full Python tool source to authenticated non-admin read-only users. ToolResponse deliberately omitted source and specs, but ToolUserResponse permitted extra fields and handlers spread a full tool model dump into the response, re-admitting omitted fields. A non-admin with a read grant can obtain another user's server-side tool source, which commonly embeds hard-coded API keys, credentials, and internal
Open WebUI: Unapproved accounts can open terminal sessions via a WebSocket auth path missing the role check
Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. From 0.8.8 until 0.11.0, the terminal WebSocket route in backend/open_webui/routers/terminals.py authenticated its own first-message JWT and never applied the verified-user role gate that get_verified_user enforces on HTTP terminal routes. An account whose role is pending, including a registered but unapproved account or an account deactivated back to pending, can open an interactive terminal session when at least one terminal server is configured and its access grants cover the account. This loses the account-approval boundary for terminal access while the