Description
vLLM is an inference and serving engine for large language models. Prior to 0.27.0, an integer overflow in blockIdx.x * 2 * d in activation_kernels.cu can cause act_and_mul_kernel to consume another batched user's input, allowing a request processed in the same inference batch to receive a partial or complete copy of another user's inference result. This issue is fixed in version 0.27.0.
Affected products
vllm: vllm
References6
- github.com/vllm-project/vllm/security/advisories/GHSA-7m6h-x95x… ExploitVendor Advisory
- github.com/vllm-project/vllm/security/advisories/GHSA-7m6h-x95x… ExploitVendor Advisory
- github.com/vllm-project/vllm/commit/451227cb3ff07989698fed982c2… Patch
- github.com/vllm-project/vllm/issues/42860 Issue TrackingPatch
- github.com/vllm-project/vllm/pull/49660 Issue TrackingPatch
- github.com/vllm-project/vllm/releases/tag/v0.27.0 Release Notes
Metrics
- CVSS 3.1
-
5.3
MEDIUM
Score source: CNA
CVSS:3.1/AV:N/AC:H/PR:N/UI:R/S:U/C:H/I:N/A:N - EPSS — exploit probability
- 0% percentile: 36 · as of 12 November 2026
- CWE
- CWE-190
Could this be exploited in your environment?
A patch closes the known hole. A penetration test shows what an attacker can still reach — with proof.