n8n before 1.123.89 contains an authentication bypass vulnerability in the Chat Trigger that allows unauthenticated attackers to skip n8n User Auth through the setup GET webhook route in embedded chat mode. Attackers can send unauthenticated GET requests to the chat webhook URL to read named session conversation history or trigger workflows with attacker-supplied input.
vLLM 0.30.0 before 0.31.0 contains a resource exhaustion vulnerability in OpenCVDynamicOpenPanguVideoBackend, which uses request-controlled num_frames without an effective ceiling. Unauthenticated attackers can send /v1/chat/completions or /tokenize requests selecting the openpangu video_backend with a huge num_frames, forcing np.linspace allocations and CPU work that OOM-kill the API server.
vLLM from 0.10.1 before 0.31.0 forwards a request-controlled mm_processor_kwargs.max_patches value to the Hugging Face processor for Cohere2 Vision models without a server-side upper bound, causing quadratic CPU work during multimodal preprocessing. Remote clients can send small image chat-completion requests with large, varying max_patches values to bypass the LRU cache, exhaust preprocessing workers, and deny service to other clients.
vLLM versions prior to 0.32.0 have a permissive default CORS configuration allowing any origin to send cross-origin requests to the HTTP server. Without an API key configured, no authentication is enforced, enabling malicious web pages to access and interact with the server if reachable by the victim's browser. This can lead to unauthorized enumeration of models and execution of chat completions, potentially consuming GPU resources and exposing model outputs. The issue is fixed in version 0.32.0.
CVE-2026-113867 is a critical vulnerability in Mozilla Thunderbird's chat feature for XMPP accounts. The flaw arises because Thunderbird did not escape attribute values when serializing outgoing XMPP stanzas. Incoming request values were echoed back in automatic replies, allowing a remote XMPP user to craft requests that cause Thunderbird to inject arbitrary stanzas into the user's connection. This enables an attacker to send messages or modify contact and subscription settings on behalf of the user without their interaction. The vulnerability was fixed in Thunderbird versions 158.0.1 and 153.5.1.
DB-GPT through 0.8.2 contains an unauthenticated remote code execution vulnerability because the /api/v1/chat/react-agent endpoint relies on a mock get_user_from_headers function that grants every caller the admin role. Unauthenticated attackers can send prompts that make the ReAct agent's unsandboxed code_interpreter and shell_interpreter tools run arbitrary Python or bash as root in the webserver container, including reaching the bundled database with default credentials.
Aix-DB 1.2.4 contains a remote code execution vulnerability that allows authenticated non-admin users to run operating system commands through the ChatBI common agent's ungated deepagents execute tool. Attackers can send crafted chat input to POST /dify/get_answer or upload documents with malicious text to run host commands, exposing datasource credentials and model API keys.
NoteGen versions up to 0.38.3 have a vulnerability where attackers with write access to the victim's sync target can manipulate file paths used in chat image analysis. This allows reading of local image files by supplying absolute paths or file:// URLs, causing the application to send local images to the AI provider. The vulnerability has a low CVSS score and requires attacker privileges and user interaction.
99AI versions up to 4.3.0 have an insecure direct object reference vulnerability that allows unauthenticated attackers to access other users' chat messages. The flaw exists because the querySingleChat function does not verify ownership of the requested chat data. Attackers can exploit this by sending a fingerprint header to obtain a visitor JWT and then sequentially iterate chatId values to retrieve message content and task data belonging to other users.
99AI versions up to and including 4.3.0 have a missing authentication vulnerability in the POST /api/chatLog/exportExcel endpoint. This flaw allows unauthenticated attackers to export all users' chat logs, including usernames, emails, prompts, and AI-generated answers, by sending crafted requests. The vulnerability has a high severity with a CVSS score of 8.7.
Open WebUI versions from 0.6.42 up to but not including 0.11.4 have a vulnerability where authenticated users can cause uncontrolled resource consumption by sending chat messages containing compressed remote image URLs. This leads to unbounded decompression of gzip-encoded responses, exhausting server memory and causing denial of service for all users. The issue is fixed in version 0.11.4.
Open WebUI versions from 0.7.0 up to but not including 0.11.4 have a cross-origin message handling vulnerability in the SyncStatsModal component. This flaw allows any website to steal a signed-in user's session token by sending unvalidated verify:chat messages. Attackers can exploit a path traversal in the chat ID to redirect authenticated requests to the authentication endpoint, thereby obtaining the victim's JWT token and gaining full API access. The vulnerability is fixed in version 0.11.4.
JeecgBoot versions through 3.9.5 have an insecure direct object reference vulnerability in the AiragBaseApiController. This flaw allows authenticated users to read AI chat variables belonging to other users by manipulating the username parameter in a specific API endpoint. The vulnerability involves sending POST requests to /airag/api/getChatVariable with a target appId, username, and variable name to access stored chat memory values in Redis. The CVSS score is 4.3, indicating a medium severity issue.
Quarkus LangChain4j versions 1.9.0 through 1.14.1 have a memory leak vulnerability in the chat-scopes WebSocket /_chat/routes endpoint. Unauthenticated remote clients can exploit this by sending repeated CONNECT frames with the same chatId, causing orphaned scopes to accumulate in memory until the JVM exits, leading to degraded server availability.
The BuddyPress Instant Chat WordPress plugin through 1.6 does not check that the sender of a chat message belongs to the conversation it is being added to, nor does it escape message content before outputting it back, allowing unauthenticated users to store arbitrary web scripts that will execute in the session of any member who later views that conversation.