https://vulnerability.circl.lu Contains only the most 10 recent entries. http://www.rssboard.org/rss-specification python-feedgen en Wed, 07 Oct 2026 09:58:52 +0000 https://vulnerability.circl.lu/vuln/fkie_cve-2026-105755 <p>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&#39;s query_key value from the caller-controlled X-Request-Id header. A concurrent request that reuses a victim&#39;s identifier can overwrite the cached query embedding so the victim&#39;s documents are scored against the attacker&#39;s query, and shared use counters can also cause a late-interaction cache-miss error. This issue is fixed in version 0.30.0.</p> <p>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&#39;s query_key value from the caller-controlled X-Request-Id header. A concurrent request that reuses a victim&#39;s identifier can overwrite the cached query embedding so the victim&#39;s documents are scored against the attacker&#39;s query, and shared use counters can also cause a late-interaction cache-miss error. This issue is fixed in version 0.30.0.</p> https://vulnerability.circl.lu/vuln/fkie_cve-2026-105755 https://vulnerability.circl.lu/vuln/ghsa-2phq-3phc-84px <p><strong>Affected:</strong> PyPI: vllm</p> <p>## Affected</p> <p>- **Ecosystem / package:** pip / `vllm` - **Affected versions:** vLLM ≤ 0.25.1 (confirmed on 0.25.1, commit [`752a3a504485`](https://github.com/vllm-project/vllm/tree/752a3a504485790a2e8491cacbb35c137339ad34)). The lower bound predates 0.25.1; maintainers can confirm how far back the flash late-interaction query cache reaches.</p> <p>## Summary</p> <p>On late-interaction `/score` and `/rerank` deployments with flash late interaction enabled (the default for supported models), the worker caches per-request query embeddings under a key derived from the **caller-controlled** `X-Request-Id` header. A second concurrent request that reuses the victim&#39;s header value replaces the victim&#39;s cached query embedding before document scoring — so the victim&#39;s documents are scored against the **attacker&#39;s** query. Because the data-parallel router pins all requests sharing a cache key to the same engine, the collision is deterministic for an attacker who reuses the victim&#39;s `X-Request-Id`. Depending on timing, one request can also consume the shared use counter and force the other request into a late-interaction cache-miss error.</p> <p>This is a remotely reachable, request-controlled cross-request integrity break on the standard scoring and reranking endpoints. It requires only that flash late interaction be enabled, which is the default for supported models.</p> <p>## Affected code</p> <p>Links pinned to the confirmed commit [`752a3a504485`](https://github.com/vllm-project/vllm/tree/752a3a504485790a2e8491c…</p> <p><strong>Affected:</strong> PyPI: vllm</p> <p>## Affected</p> <p>- **Ecosystem / package:** pip / `vllm` - **Affected versions:** vLLM ≤ 0.25.1 (confirmed on 0.25.1, commit [`752a3a504485`](https://github.com/vllm-project/vllm/tree/752a3a504485790a2e8491cacbb35c137339ad34)). The lower bound predates 0.25.1; maintainers can confirm how far back the flash late-interaction query cache reaches.</p> <p>## Summary</p> <p>On late-interaction `/score` and `/rerank` deployments with flash late interaction enabled (the default for supported models), the worker caches per-request query embeddings under a key derived from the **caller-controlled** `X-Request-Id` header. A second concurrent request that reuses the victim&#39;s header value replaces the victim&#39;s cached query embedding before document scoring — so the victim&#39;s documents are scored against the **attacker&#39;s** query. Because the data-parallel router pins all requests sharing a cache key to the same engine, the collision is deterministic for an attacker who reuses the victim&#39;s `X-Request-Id`. Depending on timing, one request can also consume the shared use counter and force the other request into a late-interaction cache-miss error.</p> <p>This is a remotely reachable, request-controlled cross-request integrity break on the standard scoring and reranking endpoints. It requires only that flash late interaction be enabled, which is the default for supported models.</p> <p>## Affected code</p> <p>Links pinned to the confirmed commit [`752a3a504485`](https://github.com/vllm-project/vllm/tree/752a3a504485790a2e8491c…</p> https://vulnerability.circl.lu/vuln/ghsa-2phq-3phc-84px