Executive brief
DocsGPT is a documentation-focused AI tool used to generate and retrieve answers from documentation. An attacker can send malicious Python code through the /api/remote endpoint that is executed on the server without authentication, potentially allowing complete compromise of the system running DocsGPT.
Technical details
The vulnerability stems from improper JSON parsing using Python's eval() function in the /api/remote endpoint (CWE-77, CWE-95). An unauthorized attacker can send arbitrary Python code that will be evaluated and executed on the server. The attack requires only network reachability to the vulnerable endpoint and no authentication. Affected versions are 0.8.1 through 0.12.0. This results in unauthenticated remote code execution with full system privileges, enabling data exfiltration, lateral movement, or system takeover.
Affected products
- Arc53 DocsGPT 0.8.1 through 0.12.0
Timeline
- 2025-02-20: disclosed: CVE-2025-0868 and GHSA-9gff-5v8w-x922 published