The flash-attention project contains a code injection vulnerability (CWE-95) in its training script due to the registration of Python's eval() function as a Hydra configuration resolver named 'eval'. This allows configuration files to execute arbitrary Python code via the ${eval:...} syntax. An attacker can exploit this by providing a malicious configuration file, resulting in arbitrary code execution when the training script processes that configuration. The vulnerability has a CVSS 3.1 base score of 7.3 (AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:L/A:L), indicating network exploitable with low attack complexity and no privileges or user interaction required. No patch or official remediation is currently documented.

Successful exploitation allows an attacker to execute arbitrary Python code on the system running the flash-attention training script by supplying a malicious configuration file. This can lead to partial confidentiality, integrity, and availability impacts as indicated by the CVSS vector. There are no known exploits in the wild at this time.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Until an official fix is available, avoid running the training script with untrusted or unaudited configuration files that could contain malicious ${eval:...} expressions. Restrict access to configuration files and validate their contents before use.