CVE-2026-115913

Summary

vLLM is an inference and serving engine for large language models. Prior to 0.32.0, the Phi-4 multimodal audio encoder in phi4mm_audio.py and phi4mm_utils.py uses eager attention by default, allowing a remote client to submit one legal long audio clip that causes MultiHeadedAttention.forward_attention to materialize and retain full attention matrices in self.attn. The scaled dot-product attention branch does not materialize these matrices. The resulting quadratic GPU-memory use can cause an out-of-memory condition and terminate EngineCore for all users. This issue is fixed in version 0.32.0.

Impact & exploitability

CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H

Affected products we track (1)

Recommendation

Apply the vendor fix in your normal patch cycle. Open any affected product above for its exact safe version.

Official patch: https://github.com/vllm-project/vllm/commit/3d52578e13454de0bb8352cac988347cdcc89928 ↗

Additional information

• NVD record
• https://github.com/vllm-project/vllm/commit/3d52578e13454de0bb8352cac988347cdcc89928Patch
• https://github.com/vllm-project/vllm/pull/60614Patch
• https://github.com/vllm-project/vllm/security/advisories/GHSA-8397-8qf6-f2gmPatch
• https://github.com/vllm-project/vllm/releases/tag/v0.32.0