CVE-2025-46570: Timing Side-Channel in vLLM Prefix Caching

# CVE-2025-46570: Timing Side-Channel in vLLM Prefix Caching

https://gcdradar.vercel.app/incidents/cve-2025-46570-timing-side-channel-in-vllm-prefix-caching · AIR-2025-0287 · closed 2025-05-28

- Lane: Patch now
- Severity: 2 (Disclosed vulnerability)
- Fix status: Fix status unknown
- First reported: 2025-05-28
- AI role: Attack on AI
- Products: vLLM
- CVE ids: CVE-2025-46570
- CVSS: 2.6

## Summary

vLLM contains a timing side-channel vulnerability linked to its chunk-based prefix caching mechanism. Differences in time to first token allow observers to infer whether specific sensitive prompts were previously cached.

## What to do

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

## Latest updates

- 2025-05-29: CVE-2025-46570: vLLM is an inference and serving engine for large language models (LLMs). (https://nvd.nist.gov/vuln/detail/CVE-2025-46570)
- 2025-05-28: CVE-2025-46570: Potential Timing Side-Channel Vulnerability in vLLM’s Chunk-Based Prefix Caching (https://github.com/advisories/GHSA-4qjh-9fv9-r85r)

## Sources

- 2025-05-28 GitHub Advisory Database: CVE-2025-46570: Potential Timing Side-Channel Vulnerability in vLLM’s Chunk-Based Prefix Caching https://github.com/advisories/GHSA-4qjh-9fv9-r85r
- 2025-05-29 NVD: CVE-2025-46570: vLLM is an inference and serving engine for large language models (LLMs). https://nvd.nist.gov/vuln/detail/CVE-2025-46570

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