Security context

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

determined-ai/determined

              main @ c1e9c6d
            

Fixes

CVEs

HIGH

Peak severity

24.3%

Coverage

Highlights

Auth Bypass: 43 prior fixes. Scrutinize any change in this area.

master/internal/api_command.go: most-fixed (6 issues). Treat as high-risk during review.

62 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.

Auth Bypass: The largest recurring risk: dozens of endpoints lacked RBAC checks or used incorrect scoping, allowing unauthorized access or modification. Fixes added checks like CanGetModel, CanEditModel, CanGetUser, CanGetWorkspace, and admin checks.

Access Control: Config policy constraints (max_slots, priority limits) were not consistently enforced across NTSC workload creation/launch paths, allowing resource limit bypass. Fixes added CheckNTSCConstraints and priority checks, but sibling paths may still miss them.

Information Disclosure: Sensitive data (task summaries, agent/slot IDs, storage credentials) was exposed to unauthorized users or leaked in logs. Fixes added obfuscation, RBAC filters, and secret masking, but other endpoints/log paths may still leak.