Auto-gapfill routes around model safety refusals with no human checkpoint
Finding PUB-F015 · 9608755858e6 · discovered by pentest-public-51classes-20260611
MEDIUMconfirmedagent-identity-abuseopus48artifact backed
- Target
- OhanaSec/jc-pentest-harness
- Target URL
- not recorded
- Run status
- completed · api-multi-model
- Framework
- agentic
- Priority score
- 0.5500
- Reachability
- unknown
- Chain depth
- 1
- PoC available
- no
- Asset
- not recorded
- Discovered
- Jun 11, 2026, 02:20 PM
Description
When the primary Hunt model refuses an attack class, gapfill silently reroutes the task to a secondary model with no human-in-the-loop checkpoint.
Validation
No validation rationale recorded.
Remediation
Add a mandatory operator approval gate before gapfill reroutes refused classes.
Evidence Boundary
This page shows persisted finding evidence, validator rationale, target context, and run linkage. Raw prompts, raw target responses, provider payloads, secrets, cookies, and authorization headers are not rendered publicly. Use the linked run page for class execution evidence, run logs, cost events, and pipeline errors.