tag
Ai-Gateway
Flowise Enterprise Authentication Bypass via Hardcoded Default JWT Secrets (CVE-2026-56271)
Flowise's enterprise passport authentication middleware signs and verifies JWTs using values pulled from environment variables (JWTAUTHTOKENSECRET, JWTREFRESHTOKENSECRET, JWTAUDIENCE, JWTISSUER). When an operator doesn't set these — an easy oversight in a…
LiteLLM Proxy Pre-Authentication SQL Injection via Error-Handling Callback (CVE-2026-42208)
LiteLLM Proxy authenticates API requests by checking that the Authorization: Bearer token starts with sk-. When a caller sends a token that does not start with sk-, that assertion fails — but instead of simply rejecting the request, the raw, unhashed token is…
LiteLLM Authentication Bypass via OIDC Userinfo Cache Key Collision (CVE-2026-35030)
LiteLLM's OIDC userinfo cache uses only the first 20 characters of the presented JWT (token[:20]) as its cache key. Two different, validly-signed JWTs can be crafted to share identical first-20-character prefixes, allowing an unauthenticated attacker to forge…
LiteLLM /config/update Broken Access Control (CVE-2026-35029)
LiteLLM's /config/update endpoint does not check the caller's role — any authenticated user holding a valid API key, not just a proxyadmin, can modify the proxy's runtime configuration. This allows registering a malicious pass-through endpoint that can be…
Authenticated Command Injection in LiteLLM MCP Test Endpoints (CVE-2026-42271)
CVE-2026-42271 is a command injection vulnerability in BerriAI LiteLLM's MCP preview/test endpoints — POST /mcp-rest/test/connection and POST /mcp-rest/test/tools/list. These endpoints accept a full MCP server configuration in the request body, including…