Browse / AI Framework and Platform Flaws
AI Framework and Platform Flaws
Classic software bugs in the libraries and platforms that run AI: memory corruption, traversal, SSRF, auth. 1,542 records. Reference: CWE.
- Denial of service 377
- Authentication and access control 256
- Path traversal and file access 192
- Memory corruption 159
- Code and command execution 133
- Server-side request forgery 103
- Information exposure 78
- Sandbox escape 76
- Cross-site scripting 68
- Other weaknesses 46
- SQL and query injection 22
- Real-world incidents 10
- Cross-site request forgery 8
- Unsafe deserialisation 8
- Data exfiltration 6
- Open WebUI has a full SSRF Vulnerability in the RAG Web Search Feature The private-address check accepts every IPv6 address because a library call fails silently, and IPv4-mapped and reserved IPv4 ranges are not blocked either.
- Open WebUI's API key endpoint restrictions bypassed via `x-api-key` header — full message processing on restricted endpoints Endpoint restrictions are enforced only for keys sent in the Authorization header, so the same key sent as x-api-key skips them.
- Open WebUI has an IDOR vulnerability in the pin_channel_message API endpoint Pin and unpin actions only check read permission on channel messages, so read-only members can change the pin state of any message.
- Open WebUI Vulnerable to Unauthenticated RAG Configuration Disclosure Open WebUI's RAG status route omits the authentication dependency that sibling routes use, so anyone can read the retrieval pipeline configuration.
- Open WebUI has Stored Cross-Site Scripting In Profile Picture Profile image data URIs with arbitrary MIME types are served from the app origin, so an embedded SVG script runs for anyone who views the image.
- Open WebUI: Low-privilege authenticated users can enumerate and stop global background tasks, causing system-wide chat disruption Any verified user can list and cancel background tasks belonging to others, because the task endpoints never check task ownership.
- Open WebUI has a Server-Side Request Forgery (SSRF) bypass in `validate_url` URL validation parses with one parser while requests uses another, so a backslash-containing URL passes the public-IP check yet connects to an internal address.
- Open WebUI vulnerable to blind server side request forgery (SSRF) via the PDF generate function HTML in the PDF export title is embedded without stripping image tags, so an img tag makes the server fetch an attacker-chosen URL.
- Open WebUI has Broken Access Control for Completions API Any API key holder can post completions into another user's chat by supplying its ID, because the server never checks who owns that chat.
- Open WebUI missing authorization check at the model update function - models from other users can be updated The model update endpoint does not verify ownership or write access, so any user can modify another user's private model.
- Open WebUI: LDAP and OAuth First-User Race Condition Allows Multiple Admin Accounts The first-user admin role is chosen before the user row is inserted, so concurrent logins on an empty instance can all receive admin.
- Open WebUI has stored XSS via unsanitized Office/Excel/DOCX file preview rendering ({@html} without DOMPurify) Office file previews are rendered as raw HTML without sanitisation, a regression of an earlier fix, so uploaded documents can run script.
- Open WebUI Vulnerable to Cross-Site Request Forgery (CSRF) via Image URL Manipulation Attacker-chosen image URLs in profile and model pictures are fetched automatically when viewed, sending authenticated GET requests to attacker endpoints.
- Open WebUI: Read-Only Users Can Toggle Note Pin Status via Incorrect Permission Check (Write via Read-Only Access) The note pin endpoint checks only read access, so a read-only collaborator can change pin state that should require write permission.
- Open WebUI has stored XSS via attacker-controlled file extension in /api/v1/audio/transcriptions A polyglot WAV/HTML file uploaded with an HTML extension is served as text/html from the cache route, so embedded scripts run in the app origin.
- Open WebUI: Cross-User File Access via Unchecked file_id in Folder Knowledge and Knowledge-Base Attach Endpoints Attach endpoints accept any file identifier without checking ownership, so an authenticated user can pull other users' files into their own retrieval context.
- Open WebUI: Missing permission check in files API allows authenticated users to list, access and delete every uploaded file File routes check only that a user is logged in and never filter by owner, so any account can list, read and delete other users' uploads.
- Open WebUI has a SSRF Bypass via HTTP Redirect Following in Web-Fetch and Image-Load Endpoints (not addressed by CVE-2025-65958) Redirects from allowed public URLs are followed without rechecking the target, reaching internal addresses through fetch and image-load paths.
- Open WebUI: Mass Assignment via FeedbackForm extra=allow Allows Feedback User ID Spoofing and Evaluation Data Manipulation Extra fields in feedback requests override the server-derived user ID, so records are attributed to other users.
- Open WebUI: Authenticated users can bypass model access control via exposed query parameter [AI-ASSISTED] An internal flag is bound from the URL query string, letting any logged-in user skip model access checks and call restricted models.
- Open WebUI has an IDOR vulnerability in the update_message_by_id API endpoint Group and DM channels only check membership when updating a message, so any member can edit messages written by other members.
- Open WebUI has an Indirect Object Reference (IDOR) in user notes The notes API does not check ownership, so any logged-in user can read other users' notes by ID.
- FlowiseAI: Dataset create+update mass-assignment allows cross-workspace dataset takeover Dataset create and update copy the entire request body onto the entity, letting a caller set workspaceId and id and move records across workspaces.
- FlowiseAI has Mass Assignment in Chatflow Update Endpoint that Allows Cross-Workspace AgentFlow Reassignment The chatflow update endpoint merges client-supplied fields such as workspaceId and deployed into the record, letting users move flows across workspaces.
- FlowiseAI has Mass Assignment in Variable Update Endpoint that Allows Cross-Workspace Resource Reassignment An authenticated user can submit server-controlled fields such as workspaceId in a variable update, moving variables into other workspaces.
- FlowiseAI: CustomTemplate create+update mass-assignment allows cross-workspace template takeover Object.assign copies a client-supplied workspaceId into a template record, letting a user take over templates in another workspace.
- FlowiseAI: Assistant create+update mass-assignment allows cross-workspace assistant takeover Client request bodies are copied directly onto the assistant record, so a caller can set the workspace identifier and access or overwrite another workspace's data.
- FlowiseAI: DatasetRow create+update mass-assignment allows cross-workspace row takeover Copying the full request body onto a database entity lets a client set its own workspace ID and move rows into another tenant's workspace.
- FlowiseAI: Authenticated Host RCE via POST /api/v1/node-custom-function and NodeVM Sandbox Escape A signed-in user posts JavaScript to a custom function endpoint with no route permission check, then escapes the NodeVM sandbox to reach host child_process.
- FlowiseAI: Vector Store No Permission Checks Vector store API routes lack permission middleware, so any logged-in user with a valid key can create, modify, upload to or delete stores they do not own.
- FlowiseAI has Mass Assignment in Assistant Update Endpoint that Allows Cross-Workspace Resource Reassignment The assistant update endpoint copies client-supplied server-owned fields such as workspaceId into the database, letting users move assistants across workspaces.
- FlowiseAI: Evaluation create+update mass-assignment allows cross-workspace evaluation takeover Object.assign copies a client-supplied workspaceId and id onto evaluations, letting an attacker move or access records belonging to other workspaces.
- FlowiseAI Exposes Basic Auth Credentials via API The basic auth check compares plaintext credentials with no rate limiting and returns distinct responses, letting attackers brute-force or enumerate logins.
- FlowiseAI: Evaluator create+update mass-assignment allows cross-workspace evaluator takeover Object.assign copies a client-supplied workspaceId into evaluator rows, letting a request move records into another workspace.
- FlowiseAI has Mass Assignment in Tool Update Endpoint that Allows Cross-Workspace Resource Reassignment The tool update endpoint merges client-supplied JSON into the database record, letting users reassign tools to other workspaces.
- n8n Has an Arbitrary File Read via Git Node The Git node's push operation passes user-supplied values as command-line options, letting a workflow editor read arbitrary server files.
- n8n Has a Cross-user Authorization Bypass in Dynamic Credential OAuth Endpoints OAuth reconnect endpoints check read permission instead of update, so read-only users can replace stored credential tokens with their own.
- n8n: HTTP Request Node Pagination Prototype Pollution to RCE A user-controlled pagination parameter in the HTTP Request node pollutes global object prototypes, which can be chained into code execution.
- n8n Has a Source Control Pull SQL Injection A malicious Data Table JSON file committed to a connected Git repository carries a crafted column name that injects SQL when an administrator runs Source Control Pull.
- n8n Has an XML Node Prototype Pollution Patch Bypass A prior fix for prototype pollution in the XML node can be circumvented by a workflow author, and combined with other nodes it reaches code execution on the host.
- Open WebUI has Stored XSS in Banner Component via Improper Sanitization Order Banner content is sanitised before Markdown conversion, so a link produced afterwards carries a javascript: URL that is rendered unescaped.
- Open WebUI: Missing `workspace.tools` Authorization Check on Tool Update Endpoint Allows Privilege Escalation to Code Execution The tool update endpoint skips the workspace permission check that tool creation enforces, so a restricted user can replace a tool's Python code and run it.
- vm2 Has a Sandbox Breakout Using Async Generator Exceptions thrown from a host object's then method inside an async generator's yield* are delivered back into the sandbox, letting code reach host constructors and run commands.
- mem0 server lacks authentication and authorization controls for its memory deletion API endpoint The mem0 DELETE /memories route has no authentication or ownership check, so any caller can erase memories for arbitrary user identifiers.
- mem0 server lacks authentication and authorization controls for its memory management API endpoints The mem0 server exposes memory update and delete endpoints without checking the caller, so unauthenticated clients can alter stored memory records.
- BentoML Dockerfile command injection via docker.base_image (sister of pending GHSA-w2pm-x38x-jp44 / CVE-2026-33744 / CVE-2026-35043) A multi-line base image value in the build config injects extra Dockerfile instructions, which then run as RUN commands during the docker build step.
- Docling's METS GBS backend is vulnerable to XML Entity Expansion (XXE) attacks Crafted XML inside a tar.gz archive triggers exponential entity expansion during parsing because entity resolution is not disabled, exhausting resources.
- Docling's JATS XML backend is vulnerable to XML Entity Expansion (XXE) attacks The JATS XML parser resolves entities without limits, so a nested entity payload expands exponentially and exhausts resources.
- LiteLLM has a sandbox escape in custom-code guardrail Bytecode-level techniques break out of the hand-rolled Python sandbox behind the custom guardrail test endpoint, running code in the proxy process.
- MLflow allows an unauthenticated remote attacker to read arbitrary files from the server's filesystem Creating a model version with a prompt tag skips source path validation, so an unauthenticated caller can register a local path that is later served as a file.