Opus 5.5 cyber safeguard blocks non-security messages after session flag

Bug Description
Opus 5.5 cyber safeguards false positives blocking legitimate security work

I'm a cybersecurity practitioner: I work in offensive security (pentest, red team) and do vulnerability research and bug bounty (YesWeHack). Claude Code is part of my daily workflow.

Since switching to Opus 5.5, my sessions get paused with "Details: [cyber]" even on trivial, non-security messages. Example: after the session was flagged once, simple messages like "yo" or "ça va ?" were blocked too, and the session stayed unusable on Opus 5.5 until I cleared it.

Two issues:

1. The session-level flag is too sticky: once triggered, any message gets blocked, including messages with no security content at all.
2. Legitimate vulnerability research and bug bounty work (authorized scope, responsible disclosure) is treated the same as malicious use, with no way to establish context or credentials.

Requests:

• Reduce false positives on non-security messages within a flagged session.
• Please open the Cyber Verification Program to Opus 5.5 (and to individual verified researchers, not only organizations). I'm happy to go through verification.

Thanks, Opus 5.5 is otherwise great and I'd like to keep using it as my main model.

Environment Info

• Platform: darwin
• Terminal: vscode
• Version: 2.1.296
• Feedback ID: 949d50ec-f707-4115-af6d-964723712d80