Fable 5 dual-use safeguard routes legitimate defensive security work off the model (false positive)

Summary

Fable 5's dual-use safeguard routed my session to Opus 4.8 during a legitimate, authorised security audit of my own developer tooling: a plugin in my own repo, on my own machine, being hardened against tampering. The flagged content was a defensive findings report using standard security vocabulary (trust boundaries, injection, secret handling, privilege). There was no malicious intent and no third-party target; this was a repo owner auditing his own code.

Impact

The routing interrupted the work and blocked access to Fable 5 for the exact task where its capability was most useful. Because defensive security and hardening work reads in the same words as offensive work, a content-only classifier cannot separate them.

Repro

1. On Fable 5, run a defensive security review of your own project (e.g. a subagent audit of a build tool's authority model and trust boundaries).
2. The findings report uses ordinary security terms (RCE, injection, exfiltration, privilege, secret handling).
3. The session is auto-routed to Opus 4.8 with a notice that the safeguards are 'intentionally broad right now'.

Request

Weight repository ownership and defensive framing, or provide a way to confirm authorised-owner context, so routine hardening on one's own environment is not gated. This is a false positive; a capability tier was lost on legitimate work.

Note

/feedback was unavailable in my setup (CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1), so filing here instead.