EUVD-2026-92845

vLLM: Flash late-interaction scoring caches query embeddings under a caller-controlled request id — cross-request integrity break and induced errors on /score and /rerank...

CVE-2026-105755

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 querykey value from the caller-controlled X-Request-Id header. A concurrent request that reuses a victim's identifier can...

CVE-2026-105755

The vLLM inference and serving engine (prior to version 0.30.0 ) is vulnerable to a cross-request integrity break in its flash late-interaction scoring at the /score and /rerank endpoints. The root cause is that the system derives the query_key value from the caller-controlled X-Request-Id header...