DeepInfer 0.1.0 contains a path traversal flaw in an unauthenticated HTTP endpoint. By manipulating the "output_dir" parameter in a GET request to the "/profile" endpoint, an attacker can cause the server to create arbitrary directories and write Chrome trace files. This can deplete a victim’s disk space or inode pool, leading to denial of service. The error messages returned by the server also reveal the existence of files, turning the vulnerability into a filesystem oracle that could assist in further reconnaissance.

The vulnerability specifically applies to DeepSeek AI’s DeepInfer 0.1.0. No other versions have been confirmed affected, and the CNA does not provide additional product or version qualifiers.

The CVSS score of 6.3 indicates medium severity, and the EPSS score is currently unavailable, which does not provide a concrete exploitation probability but suggests that the flaw may not yet have been widely targeted. The vulnerability is not listed in the CISA KEV catalog, implying no known production exploitation at this time. The attack vector is inferred to be remote, unauthenticated HTTP requests, including cross‑site requests from a legitimate operator’s browser. While the flaw cannot modify arbitrary files, the ability to create directories and exhaust resources poses a tangible operational risk for any services running DeepInfer behind exposed network interfaces.