CVE-2026-73558

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.

Weakness classification

• security-advisories@github.com: CWE-190

Primary references

• security-advisories@github.com — Patch
• security-advisories@github.com — Issue Tracking, Patch
• security-advisories@github.com — Issue Tracking, Patch
• security-advisories@github.com — Release Notes
• security-advisories@github.com — Exploit, Vendor Advisory
• 134c704f-9b21-4f2e-91b3-4a467353bcc0 — Exploit, Vendor Advisory