Security context

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

pytorch/ao

              main @ 4e42c02
            

Fixes

CVEs

CRITICAL

Peak severity

0.0%

Coverage

Highlights

Auth Bypass: 1 prior fix. Scrutinize any change in this area.

.github/workflows/ci-failure-issue.yml: most-fixed (1 issue). 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.

Auth Bypass: Deserialization of PyTorch tensors without an allowlist can lead to arbitrary code execution or unauthorized access. The fix registers WeightWithDynamicFloat8CastTensor to the allowlist, but any new tensor type added without registration could reintroduce the bypass.

Prompt Injection: The CI triage workflow feeds untrusted CI logs into Claude, which can be manipulated to trigger unintended actions. The fix adds multiple controls, but any change to the workflow could weaken them.