Fable 5 safeguards repeatedly flag legitimate load-testing/debugging of user's own app (false positive, blocks work mid-session)

Summary

Fable 5's safeguards repeatedly flag legitimate QA / load-testing / debugging work on the user's own application, aborting the session turn with:

API Error: Fable 5's safeguards flagged this message (https://www.anthropic.com/legal/aup).
Our intentionally broad safeguards allow us to deliver more capabilities faster, but can
sometimes flag legitimate coding, cybersecurity, and biology tasks. Claude Code can't
respond to this message with Fable 5.

Environment

• Claude Code CLI on Windows 11 Pro
• Model: Fable 5 (paid subscription)
• Work context: developing and testing the user's own web application on the user's own staging environment

What we were doing when it triggers

Entirely ordinary engineering work, all on systems the user owns:

• Debugging application code
• Load / concurrency testing against the user's own staging environment (Playwright browser automation simulating real user flows through the app's UI)
• Database reconciliation (read-only SELECTs to verify screen numbers match operations)

Nothing adversarial, no third-party targets, no bypassing of any protections.

Impact (as reported by the user)

• The flag has interrupted debugging and load-testing sessions repeatedly over recent weeks
• At least one previous load test could not be completed at all because of this error
• Work of the same kind used to complete without being flagged; it now gets blocked mid-session
• The user is a paying subscriber and reports significant lost time and tokens; they are requesting token compensation / usage reset through support, and asked that this report be filed on their behalf
• The user has also submitted in-product /bug reports

Ask

1. Please review the false-positive rate of these safeguards for legitimate QA / load-testing / debugging workflows (browser automation + load testing of one's own app appears to be a trigger)
2. Provide a way to continue or retry the session when a turn is flagged, instead of hard-failing the message
3. Clarify the recommended channel for compensation claims when paid sessions are repeatedly interrupted by false positives

Filed by Claude Code on behalf of the user, at the user's explicit request.