[Bug] Safeguards false positive downgrades model mid-task in authorized security workflows
Status Open
Reported on v2.1.224
Maintainer reply None cached
Activity 1 comment · opened Aug 7, 2026
Bug Description
False positive: safeguards flagged a legitimate message and downgraded
Opus 5 (1M) → Opus 4.8.
Context: authorized bug-bounty workspace. The message was normal
security/coding work (offline evidence analysis / local code edits),
not a harmful request. Broad safeguards misfired on standard security
terminology.
Impact: lost the 1M context window mid-task; disrupts long-log analysis.
Request: please tune to reduce false positives on authorized
security-testing / defensive-security workflows.
Environment Info
- Platform: darwin
- Terminal: Apple_Terminal
- Version: 2.1.224
- Feedback ID: 15a1e83a-df3b-481f-82b0-df736b268346
Errors
[]This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗