Quivr EUVD-2026-67815

Exploitation requires a valid, authenticated Quivr user account (confirmed by PR:L in the provided CVSS 4.0 vector) - no administrative privileges, elevated roles, or non-default configuration are needed. The flaw is present in default multi-user deployments of Quivr through version 0.0.322 with no additional setup required. Exploitation difficulty scales with the predictability of chat_id values: sequential integer IDs are trivially enumerable, while UUID-based IDs require a secondary information-disclosure path to obtain. Single-user Quivr deployments have no cross-user attack surface for this vulnerability. No user interaction from the victim is required.

The CVSS 4.0 base score of 8.6 accurately reflects the combination of network accessibility, low attack complexity, no attack requirements, low privilege threshold, and high confidentiality and integrity impact. The attack vector is fully network-based and requires no special attack prerequisites beyond holding a valid account, meaning the flaw is exploitable by any authenticated user against any other user in a shared deployment. EPSS data was not provided in the intelligence feed, and this CVE is absent from the CISA KEV catalog, indicating no confirmed mass exploitation. Despite the lack of a KEV listing, the IDOR pattern is operationally trivial - no memory corruption, no race condition, no specialized knowledge required - so the real-world risk for multi-tenant Quivr instances is high and should not be discounted by the absence of KEV confirmation. The confidentiality impact is particularly significant because Quivr is designed to store and query private or proprietary knowledge bases, meaning exposed chat histories may contain sensitive organizational documents or personally identifiable information ingested into the RAG pipeline.

An authenticated Quivr user queries the platform's API directly, substituting their own chat_id values with those belonging to other users - obtained via enumeration if IDs are sequential, or via other information-disclosure paths if UUIDs. Using GET /chat/{chat_id}/history, the attacker reads the full conversation transcript and any knowledge base content exposed through prior queries; using POST /chat/{chat_id}/question/answer, they inject fabricated AI responses into the victim's conversation history, silently corrupting RAG context for future queries. No public exploit code has been identified, but the attack requires only standard HTTP tooling.