CVE-2025-64496

CVE-2025-64496 Open WebUI : A code injection vulnerability in the Direct Connections feature (v0.6.224 and earlier) allows external model servers to push SSE events that execute arbitrary JavaScript in victim browsers, leading to token theft, account takeover, and potential backend RCE when combi...

CVE-2025-46719

Open WebUI vulnerability CVE-2025-46719 affects versions prior to 0.6.6. A flaw in rendering certain HTML tags in chat messages allows stored cross-site scripting (XSS) in chat transcripts, which are accessible by other users on the same server or via Open WebUI community sharing. In the user’s b...

CVE-2025-64495

Open WebUI (self-hosted offline AI platform) is affected by a Stored DOM XSS in RichTextInput when the “Insert Prompt as Rich Text” option is enabled. In versions 0.6.34 and earlier, the prompt body is parsed with marked.parse and then assigned to a temporary div’s innerHTML without sanitisation,...

CVE-2025-46571

CVE-2025-46571 affects Open WebUI prior to version 0.6.6. Low-privileged users could upload HTML files containing JavaScript via the backend endpoint /api/v1/files/, which returns a file id. An attacker could lure an admin to click a link to such a file, causing the JavaScript to execute in the a...

CVE-2026-59214

Open WebUI (self-hosted AI platform) is affected by a vulnerability described as a stored web worker XSS via Pyodide. Prior to version 0.10.0, the platform runs client-side Python with Pyodide inside a same-origin web worker, which could allow stored chat payloads using pyodide.http.pyfetch or th...

CVE-2024-30256

CVE-2024-30256 affects Open WebUI prior to version 0.1.117. The vulnerability is an authenticated blind server-side request forgery (SSRF) in the backend, specifically in the function download_file_stream() inside Open WebUI’s backend/apps/web/routers/utils.py, exploitable via the url parameter. ...

CVE-2026-45401

CVE-2026-45401 affects Open WebUI and describes an SSRF bypass: before version 0.9.5, the validate_url() check only validated the initial URL, while downstream HTTP clients (requests, aiohttp, LangChain WebBaseLoader) follow HTTP 3xx redirects by default and do not re-validate the redirected targ...

CVE-2026-44563

Open WebUI/Open WebUI’s Ollama integration vulnerability (CVE-2026-44563) affects the /api/generate, /api/embed, /api/embeddings, and /api/show endpoints. These endpoints forward a user-supplied model name to the Ollama backend without enforcing AccessGrants.has_access(), effectively bypassing mo...

CVE-2026-45398

Summary (concrete details from provided docs): Open WebUI before 0.9.5 exposes an IDOR vulnerability in the retrieval API where knowledge base collections (UUID-named) are not checked by _validate_collection_access. This allows any authenticated user who knows a private knowledge base UUID to rea...

CVE-2026-45672

Open WebUI CVE-2026-45672 affects the /api/v1/utils/code/execute endpoint, where arbitrary Python code can be executed via Jupyter for any verified user even when ENABLE_CODE_EXECUTION is false. The feature gate is not enforced at the API level, so code execution is possible despite the admin set...

CVE-2026-44552

CVE-2026-44552 affects Open WebUI. Before 0.9.0, tool_servers and terminal_servers keys in Redis were unprefixed, so when multiple instances share a Redis backend they can collide, allowing an admin on one instance to poison another’s cache and have users interact with attacker-controlled tool co...

CVE-2026-44560

Open WebUI (self-hosted offline AI platform) contains a vector-search access control flaw in the RAG retrieval path. In get_sources_from_items, non-full-context file/text collection paths can query the vector store without authorization, enabling extraction of content from files and knowledge bas...

CVE-2026-45675

Open WebUI CVE-2026-45675 describes a TOCTOU race in first-user admin role assignment for LDAP and OAuth paths prior to version 0.9.0. The signup path was fixed to insert with a default role first and upgrade if only one user remains; LDAP and OAuth paths did not receive that fix. Attack scenario...

CVE-2026-45339

Open WebUI (self-hosted offline AI platform) has a vulnerability where endpoint access restrictions on API keys could be bypassed by using the x-api-key header, even when the key was restricted from sensitive endpoints like /api/v1/messages. Prior to version 0.9.0, requests with Authorization: Be...

CVE-2026-54012

CVE-2026-54012 pertains to Open WebUI. Before version 0.9.6, a user with model-creation/update/import rights could attach forged meta.knowledge entries of type file to their model. The system then trusts these entries as authorization sources, enabling a cross-user read and deletion of private fi...

CVE-2026-44569

Open WebUI CVE-2026-44569 describes an IDOR in the channel messages management system. Before version 0.6.19, authenticated users could modify or delete any message in channels they can read because message ownership validation was missing in the backend update/delete endpoints, even though the f...

CVE-2026-45386

Technical summary (CVE-2026-45386) Open WebUI’s pin_channel_message API endpoint exposes an IDOR vulnerability in standard channels. Prior to version 0.9.5, the endpoint checks only read permission for non-admin users, allowing read-only users to pin/unpin any message in channels where they have ...

CVE-2026-45397

Open WebUI (self-hosted offline AI platform) is affected by CVE-2026-45397. The vulnerability is an information disclosure in the retrieval endpoint: GET /api/v1/retrieval/ can return live RAG configuration to unauthenticated clients. Affected component is backend/open_webui/routers/retrieval.py ...

CVE-2026-54022

Summary (grounded in provided sources): Open WebUI prior to version 0.8.11 has a logic bug in the ydoc:document:join handler: authorization is only enforced for document IDs starting with the prefix note:. The YdocManager stores documents using a normalized key where colons are replaced with unde...

CVE-2026-44554

Open WebUI (self-hosted AI) vulnerability: the POST /api/v1/retrieval/process/web endpoint accepts a user-controlled collection_name with overwrite defaulting to True, and performs no authorization check to verify write access. When overwrite is True, save_docs_to_vector_db calls VECTOR_DB_CLIENT...

CVE-2026-54017

Open WebUI vulnerability CVE-2026-54017 affects the terminal-server proxy in backend/open_webui/routers/terminals.py before version 0.9.6. An authenticated non-admin user can craft the request path to perform traversal and SSRF to the terminal server and potentially internal services. Two vectors...

CVE-2026-44551

Open WebUI vulnerability CVE-2026-44551: before version 0.9.0, the LDAP authentication endpoint does not validate non-empty passwords, allowing an unauthenticated Simple Bind on many LDAP servers. The LdapForm model accepts password: str without a minimum length, so an empty string can reach the ...

CVE-2026-44568

Summary: Open WebUI before v0.9.0 has a Stored XSS in the Pending User Overlay content. The vulnerability stems from rendering the admin-configured Pending User Overlay Content via marked.parse() inside {@html} with DOMPurify applied before markdown parsing, allowing an admin to inject JavaScript...

CVE-2026-54009

CVE-2026-54009 affects Open WebUI prior to 0.9.6. The vulnerability arises in the image_url handling path: convert_url_images_to_base64 calls get_image_base64_from_url without a user context, and get_image_base64_from_url uses Files.get_file_by_id (no ownership check) to retrieve a file by ID. Th...

CVE-2026-45399

Open WebUI CVE-2026-45399 describes a broken authorization gap in multi-user deployments: before release 0.9.0, authenticated, low-privilege users could enumerate and stop global background tasks via GET /api/tasks and POST /api/tasks/stop/{task_id}, impacting integrity and availability across us...

CVE-2026-44559

Summary (CVE-2026-44559) Open WebUI’s channel membership endpoint has an access control flaw on standard channels. Prior to version 0.9.0, GET /api/v1/channels/{id}/members only enforced membership checks for channel types ‘group’ and ‘dm’; standard (including private) channels did not perform ch...

CVE-2026-44566

Open WebUI prior to version 0.1.124 is affected by an arbitrary file upload and path traversal vulnerability. The issue occurs in the /rag/api/v1/doc upload endpoint, where the uploaded file’s name is derived from the HTTP request and is not validated or sanitized, allowing dot-segments in the fi...

CVE-2026-45385

Summary (grounded): Open WebUI (self-hosted offline AI) contains an IDOR vulnerability in the update_message_by_id API for channels of type group/dm. In these paths, the code only verifies that the caller is a channel member (is_user_channel_member) and does not confirm message ownership, enablin...

CVE-2026-45387

Open WebUI vulnerability CVE-2026-45387 affects Open WebUI (self-hosted offline AI) prior to version 0.9.5, where granting a group read access to a model could let other users view the model’s system prompt. Root cause: read-permission exposure of confidential prompt data. Impact: potential leaka...

CVE-2026-44564

Open WebUI (self-hosted offline AI platform) contains a vulnerability in the ydoc:document:update Socket.IO handler that allows read-only users to modify in-memory Yjs documents. The handler validates room membership but does not verify write permission, and read-only users join the document room...

CVE-2026-44553

Open WebUI (self-hosted offline AI) has a Socket.IO session cache vulnerability where admin role changes or user deletions are not propagated to active sessions. Prior to version 0.9.0, a user whose admin role was revoked can retain admin privileges within their existing Socket.IO session as long...

CVE-2026-45317

CVE-2026-45317 describes an application-wide CSRF vector in Open WebUI’s image handling prior to 0.9.3. An authenticated user can influence image URL rendering so that viewing a compromised image causes the user’s browser to issue GET requests to an attacker-controlled URL, potentially leaking co...

CVE-2026-54007

CVE-2026-54007 describes a cross-origin postMessage bypass in Open WebUI prior to version 0.9.6. The root cause is a chat input/submit flow in the Chat.svelte window message listener that accepts non-same-origin messages (input:prompt and action:submit) and forwards them to submitPrompt(), enabli...

CVE-2026-54019

Open WebUI CVE-2026-54019 describes an ACL bypass in Milvus multitenancy mode. Before version 0.9.6, collection-level ACL checks exist but can be bypassed when an attacker supplies user-controlled, unknown collection names, which Milvus treats as a resource_id and interpolates into an unsafe expr...

CVE-2026-45665

Open WebUI contains a Stored XSS in the Banner component due to incorrect sanitization order (DOMPurify before marked.parse). The vulnerability allows a compromised administrator to store a payload in the global banner that is rendered for all users, including the Super Admin, enabling privilege ...

CVE-2026-45671

Open WebUI vulnerability CVE-2026-45671 affects the shared-chat branch in the file authorization path. The has_access_to_file() gate unconditionally returns True for shared-chat references, ignoring the requesting user identity and the operation type. This allows any authenticated user to delete ...

CVE-2026-54018

Open WebUI (self-hosted offline AI) contains SSRF protection bypass in the Playwright Web Loader prior to version 0.9.6. The validator checks only the initial URL; Playwright follows redirects (301/302) by default, allowing an attacker-supplied URL that redirects to internal addresses (e.g., loca...

CVE-2026-44555

Open WebUI (self-hosted AI platform) has a vulnerability where a model created with base_model_id can chain to a restricted base model without validating access to that base model. Before 0.9.0, during model creation, the system does not verify the creator’s permission on the referenced base mode...

CVE-2026-44558

Open WebUI contains a vulnerability in the channel access grants path prior to version 0.9.0. The channel router does not call filter_allowed_access_grants on create or update, allowing a non-admin user who can create or own a group channel to submit arbitrary access grants (including public wild...

CVE-2026-45316

Summary (Open WebUI CVE-2026-45316): A permission check bug in the POST /api/v1/notes/{id}/pin endpoint allows read-only users to toggle a note’s is_pinned state because it checks read permission instead of write. The issue occurs in Open WebUI prior to 0.9.3 and is fixed in 0.9.3. The vulnerabil...

CVE-2026-45331

CVE-2026-45331 concerns Open WebUI’s validate_url() in backend/open_webui/retrieval/web/utils.py, where a call to validators.ipv6(ip, private=True) raises a ValidationError due to the library not implementing the private keyword for IPv6. This causes IPv6 addresses to bypass the intended filter, ...

CVE-2026-45667

Open WebUI vulnerability CVE-2026-45667: Before version 0.8.0, the unauthenticated GET /api/v1/memories/ef could trigger EMBEDDING_FUNCTION(...) and cause embedding generation, potentially incurring costs if paid providers are used. The issue is rooted in exposing a cost/resource–intensive operat...

CVE-2026-54008

Summary of CVE-2026-54008 (Open WebUI) : The vulnerable code path in backend/open_webui/utils/oauth.py::_process_picture_url validates only the initial picture_url and then fetches it with aiohttp (session.get) using default redirect-follow behavior. This enables an attacker with a valid OAuth Id...

CVE-2026-44561

CVE-2026-44561 affects Open WebUI. The vulnerability arises in the is_user_channel_member check: before 0.9.0, the code verifies ChannelMember existence but ignores is_active, so deactivated members (status 'left', is_active=False) retain full read/write access to group/DM channels via direct API...

CVE-2026-45350

Open WebUI (self-hosted AI platform) has a vulnerability in the chat_completion API prior to version 0.8.6 where user-supplied tool_ids/tool_servers are used to build a tools_dict without permission checks. This allows invoking any server tool using the server’s credentials, bypassing tool restri...

CVE-2026-44550

Open WebUI prior to 0.9.0 vulnerable to mass assignment via Pydantic extra='allow' in FolderForm. The server constructs a FolderModel by merging attacker-controlled extra fields (from form_data.model_dump(exclude_unset=True)) over a server-populated user_id, and because user_id is a real field, a...

CVE-2026-45365

Open WebUI suffers a parameter binding flaw: an internal bypass_filter parameter was exposed in the HTTP handlers for /openai/chat/completions and /ollama/api/chat via FastAPI query binding. This allowed any authenticated user to append ?bypass_filter=true and skip the ACL check, enabling access ...

CVE-2026-45396

Summary of technical details (CVE-2026-45396) Open WebUI v0.9.2 is vulnerable to mass assignment in the endpoint POST /api/v1/evaluations/feedback through a FeedbackForm that uses extra='allow'. The root cause is an insecure dictionary merge order in insert_new_feedback(), where the form data can...

CVE-2026-45402

Open WebUI CVE-2026-45402 describes a cross-user file access/overwrite vulnerability in offline Open WebUI prior to 0.9.5. Two concrete paths allow attaching a victim’s file_id without verifying ownership: (1) folder knowledge ingestion via POST /api/v1/folders/{id}/update and (2) knowledge-base ...

CVE-2025-65958

Open WebUI (self-hosted offline AI platform) is affected by a Server-Side Request Forgery (SSRF) in the /api/v1/retrieval/process/web endpoint. The vulnerability allows any authenticated user to force the server to fetch arbitrary URLs, enabling access to internal/cloud metadata endpoints (e.g., ...