AEGLIX vulnerability intelligence
CVE-2026-115913
MEDIUM · CVSS 6.5 · Undergoing Analysis
Published: 2026-10-26 · Updated: 2026-10-26 · Source: NVD
Stored catalog record
Source record retrieved 2026-10-28T10:27:20.301823+00:00. Scheduled imports handle updates; this page does not assert that the provider was queried just now.
Vulnerability overview
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.
Patch review priority
Assess exposure and vendor guidance
- No higher-priority signal is available in the current stored evidence.
FIRST EPSS: 0.30% · percentile 23.1 · score date 2026-10-27.
EPSS estimates probability of exploitation in the next 30 days. It does not measure your exposure or prove exploitation. The elevated threshold is 10%; CISA KEV takes precedence. About FIRST EPSS ↗
Severity and scoring
- CVSS score
- 6.5 · MEDIUM
- CVSS version
- 3.1
- CVSS vector
- CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H
- Weaknesses
- CWE-770
CVSS measures technical severity; it does not establish whether this vulnerability is being exploited in your environment.
Known exploitation and remediation
Not listed in the CISA catalog retrieved for this record. This does not establish that exploitation has not occurred.
Verify affected versions and patch availability in the vendor advisory before making changes.
Affected products and versions
NVD has not supplied affected-version configurations for this record. Check the vendor references below.