Librechat CVE-2026-31950

While the CVSS score of 5.3 (medium) reflects limited attack complexity (AC:H) and local scope, the real-world risk is moderate-to-high when contextual factors are considered. The vulnerability requires authentication (PR:L), limiting exposure to users with valid accounts; however, in multi-tenant SaaS deployments or shared LibreChat instances, account registration may be trivial. The confidentiality impact is high (C:H) - attackers gain direct access to other users' conversations, potentially exposing sensitive data, API keys, personal information, or proprietary discussions. The attack complexity is high (AC:H), which typically accounts for the need to guess or enumerate valid stream IDs; however, stream IDs may be predictable or enumerable if not cryptographically random. No KEV status or public exploit code is confirmed in the provided data, suggesting limited real-world weaponization at analysis time. The CVSS vector accurately captures the profile: remote delivery, authentication required, high confidentiality impact, no integrity or availability compromise. Prioritize patching in multi-tenant or shared deployments where unauthorized information disclosure poses regulatory or reputational risk.

An authenticated user in a multi-tenant LibreChat deployment discovers or enumerates stream IDs via timing analysis, error messages, or API observation. The attacker crafts a request to /api/agents/chat/stream/[victim_stream_id] and establishes an SSE connection, receiving a real-time stream of another user's chat messages and AI responses without ownership verification. The attacker passively monitors the stream to extract sensitive information such as API keys, credentials, or proprietary business logic discussed in the conversation. No active exploit code or in-the-wild attacks are documented in the provided intelligence; however, the simplicity of the attack (authentication + stream ID enumeration) makes it likely to be discovered and exploited once publicly disclosed.