Response corruption via predictable prefix-cache hash collisions
Published Feb 7, 2025 · Updated Feb 12, 2025
Cache integrity failure in vLLM before 0.7.2 allows remote authenticated users to corrupt subsequent responses via colliding prompts. PrefixCachingBlock.hash_block_tokens seeds the first-block tuple with Python's hash(None), which is constant under Python 3.12, so distinct token sequences can share a cache key. Exploitation requires prefix caching, knowledge of another prompt, a crafted colliding prompt, and a later request that reuses the poisoned entry; the bounded impact is altered model output.
Summary
What happened
Cache integrity failure in vLLM before 0.7.2 allows remote authenticated users to corrupt subsequent responses via colliding prompts. PrefixCachingBlock.hash_block_tokens seeds the first-block tuple with Python's hash(None), which is constant under Python 3.12, so distinct token sequences can share a cache key. Exploitation requires prefix caching, knowledge of another prompt, a crafted colliding prompt, and a later request that reuses the poisoned entry; the bounded impact is altered model output.
The record
- CVE
- CVE-2025-25183
- Published
- Feb 7, 2025
- Updated
- Feb 12, 2025
- Vendor
- vLLM Project
- Product
- vLLM
- Classifications
- CWE-354, T1565.003
- Attack vector
- network
- Privileges
- authenticated
Timeline
How it unfolded
- Feb 7, 2025CVE publishedPublication date reported by the CVE source.
- Feb 12, 2025Record updatedLatest update available in the CVE record.
Exploitability
Present is not the same as exploitable
Compare your product and version with the public record. A matching version still requires validation against your environment.
Is a vulnerable build present?
Compare these published version ranges with your installed build and any vendor patches.
- Affected versionversion=< 0.7.2
What conditions does exploitation require?
What is affected?
Published CVSS scores
CVSS describes severity. EPSS estimates exploitation probability.
Attacks
What attackers are doing with it
Daily unique IPs observed by Shadowserver honeypots for known exploited vulnerabilities (KEVs). Missing observations do not establish an absence of attacks.
Weakness, pattern, technique
Public exploit references
No public exploit references are available in this record.
Labels summarize the accepted research assessment. They do not indicate a test against your environment.
Technologies
Your stack
See the directory against your own environment.
Your stack
Check the software in your environment
Book a demo to see how Hinoki identifies affected software and validates exploitability in your environment.
Book a demo