CVE-2025-46560: vLLM vulnerable to denial of service via quadratic time complexity

# CVE-2025-46560: vLLM vulnerable to denial of service via quadratic time complexity

https://gcdradar.vercel.app/incidents/cve-2025-46560-vllm-vulnerable-to-denial-of-service-via-quad · AIR-2025-0278 · closed 2025-04-29

- Lane: Patch now
- Severity: 2 (Disclosed vulnerability)
- Fix status: Fix status unknown
- First reported: 2025-04-29
- AI role: Attack on AI
- Products: vLLM
- CVE ids: CVE-2025-46560
- CVSS: 6.5

## Summary

A vulnerability in vLLM's multimodal tokenizer input processor causes quadratic time complexity leading to denial of service. The issue allows malicious actors to trigger resource exhaustion via crafted inputs.

## What to do

No action recorded; read the sources below and the vendor's advisory.

## Latest updates

- 2025-04-30: CVE-2025-46560: vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. (https://nvd.nist.gov/vuln/detail/CVE-2025-46560)
- 2025-04-29: CVE-2025-46560: vLLM: Quadratic Time Complexity in Input Token Processing​ leads to denial of service (https://github.com/advisories/GHSA-vc6m-hm49-g9qg)

## Sources

- 2025-04-29 GitHub Advisory Database: CVE-2025-46560: vLLM: Quadratic Time Complexity in Input Token Processing​ leads to denial of service https://github.com/advisories/GHSA-vc6m-hm49-g9qg
- 2025-04-30 NVD: CVE-2025-46560: vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. https://nvd.nist.gov/vuln/detail/CVE-2025-46560

Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.