CVE-2026-116180
vLLM from 0.8.5 before 0.31.0 has a denial of service flaw
- Risk score
- 46/100
- CVSS
- 7.7 v3.1
- Published
- 25 Oct 2026
Open in the live forecast Today's forecast
What to know
- What it is
- vLLM from 0.8.5 before 0.31.0 has a denial of service flaw. LogprobsTensors.empty_cpu() in vllm/v1/outputs.py allocates an uninitialized int32 tensor, and the V1 model runner leaves prompt positions reported as computed by a KV connector unwritten. A client sharing the engine can send prompt_logprobs (or echo with log…
- Affected
- See advisory
- Fix
- Upgrade the product to 0.31.0 or later.
- Exploitation
- No known exploitation
Why it is Yellow
- CVSS 7.7 high
References
Original description
vLLM from 0.8.5 before 0.31.0 has a denial of service flaw. LogprobsTensors.empty_cpu() in vllm/v1/outputs.py allocates an uninitialized int32 tensor, and the V1 model runner leaves prompt positions reported as computed by a KV connector unwritten. A client sharing the engine can send prompt_logprobs (or echo with logprobs) against a connector-cached prefix, making tokenizer decode raise OverflowError, which crashes EngineCore and drops all co-tenant requests.
Last activity 25 Oct 2026. Patch Forecast updates every hour.