Security context
What an agent needs to avoid regressing past fixes and find the next vuln in this repo.
NVIDIA/garak
main @ af5564f
8
Fixes
0
CVEs
HIGH
Peak severity
50.0%
Coverage
Highlights
Denial of Service: 4 prior fixes. Scrutinize any change in this area.
garak/generators/openai.py: most-fixed (2 issues). Treat as high-risk during review.
1 high-severity fix in this history; regressions here are high-impact.
Recurring patterns
The bug types that recur here, drawn from past fixes, not open vulnerabilities.
Remote Code Execution: Loading untrusted third-party transformer models without explicitly validating the trust_remote_code parameter allows arbitrary code execution during weights initialization. This is particularly dangerous for an LLM security assessment tool that routinely processes external models.
Denial of Service: Concurrency structures utilizing multiprocessing.Pool can hang permanently if worker threads encounter unpicklable exceptions (such as custom Authentication or PermissionDenied errors) across process boundaries, or fall into infinite backoff loops when handling unexpected null API responses.
SQL Injection: Generating digests from scan reports using unsafe string formatting (f-strings) in database execution blocks allows arbitrary SQL injection. Garak must strictly utilize parameterized queries when saving or reading scan metadata.