Unsloth Studio Flaw Turns Routine Model Inspection Into Code Execution
Summary
A vulnerability in Unsloth Studio has been patched, which previously allowed malicious AI models to execute arbitrary Python code when inspected. This was possible through the 'trust_remote_code' setting, turning routine model inspection into a potential code execution risk.
IFF Assessment
The vulnerability allowed malicious code execution via AI model inspection, posing a direct threat to systems running these models.
Severity
The vulnerability, allowing arbitrary code execution with the 'trust_remote_code' setting, presents a significant attack vector and impact, warranting a high CVSS score.
Defender Context
This highlights the risks associated with loading and inspecting AI models from untrusted sources, especially when remote code execution is enabled. Defenders should ensure that such settings are carefully managed and that AI model provenance is verified.