The flaw exists in the EEGModuleMixin._from_pretrained function of braindecode’s Hub Integration. The filename parameter is not properly validated, allowing an attacker to supply a crafted path that escapes the intended directory. If exploited, the malicious path can be used to read any file that the process can access, potentially exposing configuration files, secrets, or other sensitive data. The impact therefore is the ability to disclose information that should not be publicly available.

Braindecode libraries released up to version 1.8.2 are affected. The patch that fixes the vulnerability is included in release 2.0.0. Administrators should verify that the component braindecode (cpe:2.3:a:braindecode:braindecode) is updated to at least v2.0.0 or later.

The CVSS score of 5.3 indicates moderate severity; EPSS data is not available, and the vulnerability has not yet been listed in the CISA KEV catalog. The attack vector is remote – an attacker can trigger the flaw by invoking the vulnerable function from any exposed interface. If the code runs with elevated privileges, the risk can increase to higher impact, potentially enabling broader compromise. The lack of publicly documented exploits suggests the risk is moderate, but the path traversal flaw is common enough that it warrants prompt mitigation.