CVE-2025-3263: 
NixOS vulnerability analysis and mitigation

CVE-2025-3263 is a Regular Expression Denial of Service (ReDoS) vulnerability in the Hugging Face Transformers library, specifically within the get_configuration_file() function of the transformers.configuration_utils module. The vulnerability affects versions starting from 4.49.0 and is resolved in version 4.51.0. It was disclosed on July 7, 2025, and carries a CVSS v3.0 base score of 5.3 (Medium) (GitHub Advisory, Red Hat Bugzilla).

The root cause is an inefficient regular expression pattern config\.(.*)\.json used in the get_configuration_file() function, classified as CWE-1333 (Inefficient Regular Expression Complexity). The greedy (.*) quantifier between two literal dot-escaped segments enables catastrophic backtracking when processing specially crafted input strings, causing exponential CPU consumption. An unauthenticated remote attacker can trigger this condition by supplying a malicious configuration file name string to any application endpoint that invokes this function, requiring no privileges or user interaction. A proof-of-concept is publicly available via the huntr.com bug bounty platform (huntr PoC, GitHub Advisory).

Successful exploitation results in excessive CPU consumption on the affected host, leading to model serving disruption, resource exhaustion, and significantly increased application latency. The impact is limited to availability — there is no confidentiality or integrity impact. Applications built on the Hugging Face Transformers library (versions 4.49.0 to below 4.51.0) that expose configuration file resolution to external input are at risk of denial of service (GitHub Advisory, Red Hat Bugzilla).

A proof-of-concept exploit is publicly available on huntr.com, submitted through the huntr bug bounty program. There is no current evidence of in-the-wild exploitation or threat actor attribution. The EPSS score is approximately 0.042% (Feedly data) to 0.096% (GitHub Advisory), placing it in the lower percentiles for near-term exploitation likelihood. The vulnerability is not listed in the CISA Known Exploited Vulnerabilities (KEV) catalog (huntr PoC, GitHub Advisory).

1. Identify target: Locate an application or API endpoint that uses Hugging Face Transformers versions 4.49.0 to 4.50.x and accepts user-controlled input that influences configuration file name resolution (e.g., model loading endpoints).
2. Craft malicious input: Construct a string designed to trigger catastrophic backtracking in the regex config\.(.*)\.json. For example, a long string like config. followed by many repeated characters and a non-matching suffix (e.g., config.aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa!) causes the regex engine to explore exponentially many paths.
3. Submit the payload: Send the crafted string to the target endpoint so it is processed by get_configuration_file() in transformers.configuration_utils.
4. Observe denial of service: The regex engine enters catastrophic backtracking, consuming excessive CPU cycles and causing the application to become unresponsive or severely degraded, achieving a denial-of-service condition (huntr PoC, GitHub Advisory).

• Application Performance: Sudden and sustained CPU spikes on servers running Hugging Face Transformers-based applications, particularly during model configuration loading operations.
• Logs: Application logs showing repeated or long-running calls to get_configuration_file() in transformers/configuration_utils.py; timeout errors or unresponsive model-serving endpoints coinciding with unusual input strings.
• Network: Repeated requests to model-loading or configuration endpoints with abnormally long or malformed config_name parameters containing patterns like config. followed by many repeated characters.
• Process: Python worker processes consuming near-100% CPU for extended periods without completing normal model loading tasks.

The primary remediation is to upgrade the Hugging Face Transformers library to version 4.51.0 or later, which resolves the vulnerable regex pattern (pip install transformers>=4.51.0) (GitHub Advisory, GitHub Commit). As interim mitigations, implement input validation to sanitize or restrict configuration file name inputs before they reach the library, apply timeout mechanisms for regex processing, and monitor application CPU usage for anomalies. Organizations should prioritize upgrading any internet-facing services that accept user-controlled model or configuration names.

Red Hat tracked the vulnerability via their security response process (Bugzilla bug 2376773) and assigned it medium severity. The GitHub Advisory Database published and reviewed the advisory on July 7–8, 2025. No notable public researcher commentary or significant social media discussion has been identified beyond standard vulnerability aggregator coverage.