GHSA-RM76-4MRF-V9R8
Vulnerability from github – Published: 2025-02-06 20:00 – Updated: 2025-07-02 14:20Summary
Maliciously constructed prompts can lead to hash collisions, resulting in prefix cache reuse, which can interfere with subsequent responses and cause unintended behavior.
Details
vLLM's prefix caching makes use of Python's built-in hash() function. As of Python 3.12, the behavior of hash(None) has changed to be a predictable constant value. This makes it more feasible that someone could try exploit hash collisions.
Impact
The impact of a collision would be using cache that was generated using different content. Given knowledge of prompts in use and predictable hashing behavior, someone could intentionally populate the cache using a prompt known to collide with another prompt in use.
Solution
We address this problem by initializing hashes in vllm with a value that is no longer constant and predictable. It will be different each time vllm runs. This restores behavior we got in Python versions prior to 3.12.
Using a hashing algorithm that is less prone to collision (like sha256, for example) would be the best way to avoid the possibility of a collision. However, it would have an impact to both performance and memory footprint. Hash collisions may still occur, though they are no longer straight forward to predict.
To give an idea of the likelihood of a collision, for randomly generated hash values (assuming the hash generation built into Python is uniformly distributed), with a cache capacity of 50,000 messages and an average prompt length of 300, a collision will occur on average once every 1 trillion requests.
References
- https://github.com/vllm-project/vllm/pull/12621
- https://github.com/python/cpython/commit/432117cd1f59c76d97da2eaff55a7d758301dbc7
- https://github.com/python/cpython/pull/99541
{
"affected": [
{
"package": {
"ecosystem": "PyPI",
"name": "vllm"
},
"ranges": [
{
"events": [
{
"introduced": "0"
},
{
"fixed": "0.7.2"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2025-25183"
],
"database_specific": {
"cwe_ids": [
"CWE-354"
],
"github_reviewed": true,
"github_reviewed_at": "2025-02-06T20:00:05Z",
"nvd_published_at": "2025-02-07T20:15:34Z",
"severity": "LOW"
},
"details": "### Summary\n\nMaliciously constructed prompts can lead to hash collisions, resulting in prefix cache reuse, which can interfere with subsequent responses and cause unintended behavior.\n\n### Details\n\nvLLM\u0027s prefix caching makes use of Python\u0027s built-in hash() function. As of Python 3.12, the behavior of hash(None) has changed to be a predictable constant value. This makes it more feasible that someone could try exploit hash collisions.\n\n### Impact\n\nThe impact of a collision would be using cache that was generated using different content. Given knowledge of prompts in use and predictable hashing behavior, someone could intentionally populate the cache using a prompt known to collide with another prompt in use. \n\n### Solution\n\nWe address this problem by initializing hashes in vllm with a value that is no longer constant and predictable. It will be different each time vllm runs. This restores behavior we got in Python versions prior to 3.12.\n\nUsing a hashing algorithm that is less prone to collision (like sha256, for example) would be the best way to avoid the possibility of a collision. However, it would have an impact to both performance and memory footprint. Hash collisions may still occur, though they are no longer straight forward to predict.\n\nTo give an idea of the likelihood of a collision, for randomly generated hash values (assuming the hash generation built into Python is uniformly distributed), with a cache capacity of 50,000 messages and an average prompt length of 300, a collision will occur on average once every 1 trillion requests.\n\n### References\n\n* https://github.com/vllm-project/vllm/pull/12621\n* https://github.com/python/cpython/commit/432117cd1f59c76d97da2eaff55a7d758301dbc7\n* https://github.com/python/cpython/pull/99541",
"id": "GHSA-rm76-4mrf-v9r8",
"modified": "2025-07-02T14:20:34Z",
"published": "2025-02-06T20:00:05Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/vllm-project/vllm/security/advisories/GHSA-rm76-4mrf-v9r8"
},
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2025-25183"
},
{
"type": "WEB",
"url": "https://github.com/python/cpython/pull/99541"
},
{
"type": "WEB",
"url": "https://github.com/vllm-project/vllm/pull/12621"
},
{
"type": "WEB",
"url": "https://github.com/python/cpython/commit/432117cd1f59c76d97da2eaff55a7d758301dbc7"
},
{
"type": "WEB",
"url": "https://github.com/vllm-project/vllm/commit/73b35cca7f3745d07d439c197768b25d88b6ab7f"
},
{
"type": "WEB",
"url": "https://github.com/pypa/advisory-database/tree/main/vulns/vllm/PYSEC-2025-62.yaml"
},
{
"type": "PACKAGE",
"url": "https://github.com/vllm-project/vllm"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:H/PR:L/UI:R/S:U/C:N/I:L/A:N",
"type": "CVSS_V3"
}
],
"summary": "vLLM uses Python 3.12 built-in hash() which leads to predictable hash collisions in prefix cache"
}
Sightings
| Author | Source | Type | Date |
|---|
Nomenclature
- Seen: The vulnerability was mentioned, discussed, or observed by the user.
- Confirmed: The vulnerability has been validated from an analyst's perspective.
- Published Proof of Concept: A public proof of concept is available for this vulnerability.
- Exploited: The vulnerability was observed as exploited by the user who reported the sighting.
- Patched: The vulnerability was observed as successfully patched by the user who reported the sighting.
- Not exploited: The vulnerability was not observed as exploited by the user who reported the sighting.
- Not confirmed: The user expressed doubt about the validity of the vulnerability.
- Not patched: The vulnerability was not observed as successfully patched by the user who reported the sighting.