[Bug] Fable 5 safeguards blocking non-technical disagreement statements in security conversations

Bug Description
False positive: Fable 5 safeguards blocked a message with no technical content. Request ID: req_011Cdu5atug3xinQip2nFADT Model: Fable 5, Claude Code CLI (macOS) What happened: I asked whether Fable 5 can at least be used to plan security settings. Claude answered that defensive work on one's own systems is not affected by the additional safeguards. I replied, verbatim and in full: "es ist leider nicht so wie du denkst" ("unfortunately it's not the way you think"). That eight-word message was blocked. Why this is a problem: 1. The blocked message contained no request, no code, no target system, no technical content whatsoever — it was a one-line disagreement. 2. The block occurred in a meta-conversation about the safeguards. Talking about the limits triggers the limits, so a user cannot even establish what is possible. 3. It ironically proved the point it blocked: the model had just claimed such work is unaffected. 4. It forced a mid-session model switch, losing working context. Context: solo developer doing defensive ops work on my own infrastructure — FileVault/firewall config on my own Macs, sudo and SSH key handling, auth guards in my own API. Nothing offensive, nothing third-party. The classifier appears to fire on topic and conversational shape rather than on intent or target. Please review.

Environment Info

• Platform: darwin
• Terminal: Apple_Terminal
• Version: 2.1.226
• Feedback ID: 8de834ce-1121-4214-8229-8f01c2162952

Errors

[]