vLLM is an inference and serving engine for large language models. Prior to 0.30.0, flash late-interaction scoring at the /score and /rerank endpoints derives each worker's query_key value from the caller-controlled X-Request-Id header. A concurrent request that reuses a victim's identifier can overwrite the cached query embedding so the victim's documents are scored against the attacker's query, and shared use counters can also cause a late-interaction cache-miss error.
This issue is fixed in version 0.30.0.
Attack Parameters
Attack Vector
Network
Can be exploited remotely
Attack Complexity
High
Difficult to exploit
Privileges Required
Low
Basic privileges needed
User Interaction
None
No user interaction needed
Impact Assessment
Confidentiality
None
No data leak
Integrity
Low
Partial data modification
Availability
Low
Partial disruption
CVSS Vector v3.1
Weakness Type (CWE)
Vulnerable Products
vllm-project:vllm
References 4
https://github.com/vllm-project/vllm/security/advisories/GHSA-2phq-3phc-84px
https://github.com/vllm-project/vllm/security/advisories/GHSA-2phq-3phc-84px
https://github.com/vllm-project/vllm/pull/51445
https://github.com/vllm-project/vllm/pull/51445
https://github.com/vllm-project/vllm/commit/ee17d0d869203ef9a35ad73358a4987bba1…
https://github.com/vllm-project/vllm/commit/ee17d0d869203ef9a35ad73358a4987bba14b1fc
https://github.com/vllm-project/vllm/releases/tag/v0.30.0
https://github.com/vllm-project/vllm/releases/tag/v0.30.0