[Bug] Fable 5 safeguards falsely blocking meta-questions about safety decisions

Bug Description
Subject: False positive from Fable 5 safeguards on a routine coding conversation

Request ID: req_011CdJTb31Gg8XAQmp4YedAY

What happened: During a normal Claude Code session, Fable 5's safeguards blocked one of my messages and cited the AUP. The session was ordinary app development work: I was debugging a Bluetooth-related video playback bug (GitHub issue #292) in an Unreal Engine mobile VR project, reviewing a C++ diff, and routing the same diff to two other models for a second opinion.

The message that got flagged: I asked the assistant why my earlier message had tripped the Fable 5 safeguards. That question, phrased with words like "safeguards," "what did I do wrong," and "flagged," was itself blocked. The error text even says this "sometimes happens with safe, normal conversations."

Why I think it's a false positive: Nothing in the conversation touches the AUP. The whole thread is Android/iOS lifecycle handling, media player pause and resume logic, and crash triage on a shipping app. The earlier context did carry security-adjacent vocabulary common to mobile development (crash reporter, signal handler, JNI, secure storage, symbolication), and I suspect that vocabulary, combined with a meta-question that repeated the words "safeguards" and "flagged," is what pushed the input classifier over the line. None of it is a real policy concern.

Impact: I couldn't get an answer on Fable 5 and had to switch models to continue. The block interrupted legitimate debugging work with no unsafe request anywhere in the session.

Ask: Please review this request ID as a false positive and use it to tune the input classifier. Asking Claude to explain its own safety behavior should not itself be treated as a violation.

Environment Info

• Platform: win32
• Terminal: Orca
• Version: 2.1.218
• Feedback ID: 5c3e44df-0591-4c9a-9168-17190f71c7d0

Errors

[]