CVE-2025-3264 is a Regular Expression Denial of Service (ReDoS) vulnerability identified in the Hugging Face Transformers library, specifically within the `get_imports()` function located in the `dynamic_module_utils.py` file. This vulnerability affects version 4.49.0 of the library and was addressed in version 4.51.0. The root cause lies in the use of an inefficient regular expression pattern `\s*try\s*:.*?except.*?:` designed to filter out try/except blocks from Python code. This pattern is susceptible to catastrophic backtracking when processing specially crafted input strings, which can cause excessive CPU consumption and lead to denial of service conditions. The vulnerability does not directly compromise confidentiality or integrity but impacts availability by exhausting computational resources. Potential exploitation scenarios include disruption of remote code loading processes, resource exhaustion during model serving, interference with supply chain operations, and disruption of development pipelines that rely on the vulnerable library. The vulnerability is remotely exploitable without authentication or user interaction, increasing its risk profile. The CVSS 3.0 base score is 5.3 (medium severity), reflecting the network attack vector, low attack complexity, no privileges required, no user interaction, and impact limited to availability. No known exploits are currently reported in the wild, but the widespread use of Hugging Face Transformers in AI/ML workflows makes this a relevant concern for organizations leveraging these tools.

For European organizations, the impact of this vulnerability can be significant, especially for those heavily invested in AI and machine learning workflows that utilize the Hugging Face Transformers library. Organizations providing AI model serving or inference as a service could face service disruptions due to resource exhaustion caused by maliciously crafted inputs exploiting this ReDoS vulnerability. This could lead to downtime, degraded performance, and potential cascading effects on dependent applications and services. Development teams relying on automated pipelines that incorporate the vulnerable library might experience delays or failures, impacting release cycles and operational efficiency. Additionally, supply chain risks arise if malicious actors exploit this vulnerability to disrupt or manipulate AI model deployment processes. While the vulnerability does not directly expose sensitive data or allow code execution, the availability impact could affect critical AI-driven services in sectors such as finance, healthcare, telecommunications, and public services across Europe. The medium severity rating suggests the threat is manageable but requires timely remediation to avoid operational risks.

European organizations should prioritize upgrading the Hugging Face Transformers library to version 4.51.0 or later, where the vulnerability is fixed. In environments where immediate upgrade is not feasible, organizations can implement input validation and sanitization to detect and block suspiciously complex or malformed inputs that could trigger the ReDoS condition. Monitoring CPU and memory usage patterns in AI model serving infrastructure can help detect anomalous resource consumption indicative of exploitation attempts. Incorporating rate limiting and request throttling mechanisms on endpoints that process user-supplied code or model inputs can reduce the risk of resource exhaustion. Additionally, organizations should review and harden their development pipelines to ensure that untrusted or external code is not processed without adequate scrutiny. Employing static analysis tools to detect inefficient regular expressions and potential ReDoS patterns in custom code can prevent similar vulnerabilities. Finally, maintaining an up-to-date inventory of AI/ML dependencies and integrating vulnerability scanning into CI/CD workflows will help promptly identify and remediate such issues.