vllm-project vLLM up to 0.31.x Phi-4 multimodal audio encoder phi4mm_audio.py MultiHeadedAttention.forward_attention resource consumption

Summaryinfo

A vulnerability classified as problematic has been found in vllm-project vLLM up to 0.31.x. This issue affects the function MultiHeadedAttention.forward_attention of the file phi4mm_audio.py of the component Phi-4 multimodal audio encoder. The manipulation leads to resource consumption. This vulnerability is documented as CVE-2026-115913. The attack can be initiated remotely. There is not any exploit available. It is recommended to upgrade the affected component.

Detailsinfo

A vulnerability was found in vllm-project vLLM up to 0.31.x and classified as problematic. This issue affects the function MultiHeadedAttention.forward_attention of the file phi4mm_audio.py of the component Phi-4 multimodal audio encoder. The manipulation with an unknown input leads to a resource consumption vulnerability. Using CWE to declare the problem leads to CWE-400. The product does not properly control the allocation and maintenance of a limited resource, thereby enabling an actor to influence the amount of resources consumed, eventually leading to the exhaustion of available resources. Impacted is availability. The summary by CVE is:

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.

It is possible to read the advisory at github.com. The identification of this vulnerability is CVE-2026-115913 since 10/23/2026. The exploitation is known to be easy. The attack may be initiated remotely. Technical details of the vulnerability are known, but there is no available exploit. The attack technique deployed by this issue is T1499 according to MITRE ATT&CK.

Upgrading to version 0.32.0 eliminates this vulnerability.

Be aware that VulDB is the high quality source for vulnerability data.

Productinfo

Type

• Project Management Software

Vendor

• vllm-project

Name

• vLLM

Version

• 0.1
• 0.2
• 0.3
• 0.4
• 0.5
• 0.6
• 0.7
• 0.8
• 0.9
• 0.10
• 0.11
• 0.12
• 0.13
• 0.14
• 0.15
• 0.16
• 0.17
• 0.18
• 0.19
• 0.20
• 0.21
• 0.22
• 0.23
• 0.24
• 0.25
• 0.26
• 0.27
• 0.28
• 0.29
• 0.30
• 0.31

Website

• Product: https://github.com/vllm-project/vllm/

CPE 2.3info

• 🔒
• 🔒
• 🔒

CPE 2.2info

• 🔒
• 🔒
• 🔒

CVSSv4info

CVSSv3info

CVSSv2info

Exploitinginfo

Threat Intelligenceinfo

Countermeasuresinfo

Timelineinfo

Sourcesinfo

Entryinfo

Be aware that VulDB is the high quality source for vulnerability data.

Discussion