CyberRota Analysis
AI-GeneratedThe vulnerability exists in vLLM versions up to 0.29.0, where the Triton _bincount_kernel improperly handles prompt token IDs, leading to potential memory corruption due to lack of bounds checking against the vocabulary size. This can allow attackers to exploit multimodal audio requests, resulting in out-of-bounds writes that may disrupt the sampler state of concurrent requests and modify repetition penalty behavior. Organizations using affected versions of vLLM should prioritize patching this vulnerability to mitigate the risk of exploitation.
Public Exploit Signal
A public exploit, PoC, GitHub repository or Metasploit reference was detected for this CVE.
Note: these links are listed for security research and verification purposes only.
Original NVD Description
vLLM through 0.29.0 contains a memory corruption vulnerability in the Triton _bincount_kernel where prompt token IDs index the penalty prompt-presence bitset without bounds checking against vocabulary size. Attackers can submit multimodal audio requests with tokens equal to vocabulary size, causing out-of-bounds writes that corrupt concurrent requests' sampler state and alter repetition penalty behavior.
Related CVEs
Other vulnerabilities affecting the same vendor(s)