CVE-2026-115913
vLLM: Phi-4 multimodal audio encoder materializes and retains full attention matrices: a single max-length audio OOM-crashes the vLLM engine
CVSS Score
6.5
EPSS Score
0.0%
EPSS Percentile
0th
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.
| CWE | CWE-770 |
| Vendor | vllm-project |
| Product | vllm |
| Published | Oct 26, 2026 |
Stay Ahead of the Next One
Get instant alerts for vllm-project vllm
Be the first to know when new medium vulnerabilities affecting vllm-project vllm are published — delivered to Slack, Telegram or Discord.
Get Free Alerts →
Free · No credit card · 60 sec setup
CVSS v3 Breakdown
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H Attack Vector
Network
Attack Complexity
Low
Privileges Required
Low
User Interaction
None
Scope
Unchanged
Confidentiality
None
Integrity
None
Availability
High
Affected Versions
vllm-project / vllm
< 0.32.0
References
github.com: https://github.com/vllm-project/vllm/security/advisories/GHSA-8397-8qf6-f2gm github.com: https://github.com/vllm-project/vllm/pull/60614 github.com: https://github.com/vllm-project/vllm/commit/3d52578e13454de0bb8352cac988347cdcc89928 github.com: https://github.com/vllm-project/vllm/releases/tag/v0.32.0