Security context

What an agent needs to avoid regressing past fixes and find the next vuln in this repo.

intel/neural-compressor

              main @ debb577
            

Fixes

CVEs

CRITICAL

Peak severity

Highlights

Code Execution: 2 prior fixes. Scrutinize any change in this area.

neural_solution/frontend/fastapi/main_server.py: most-fixed (2 issues). Treat as high-risk during review.

8 high-severity fixes in this history; regressions here are high-impact.

Recurring patterns

The bug types that recur here, drawn from past fixes, not open vulnerabilities.

Code Execution: The use of the built-in `eval` function on configuration keys and operation types allowed arbitrary python code execution. Developers should enforce safe regex-based parsing or literal ast evaluations instead.

Deserialization of Untrusted Data: Standard `torch.load` of untrusted pickles allows RCE. The library had to transition to strict blocklists and enforcing weights_only deserialization to mitigate model-based exploit chains.

Command Injection: A lack of input validation on task requests allowed attackers to inject shell-metacharacters into command arguments constructed for backend tasks. Strict input blocklisting and safe list-based execution must be mandated.