[Bug] Fable 5 silently downgraded to Opus 4.8 for defensive security work without user indication or override option

Bug Description
Fable 5 silently re-routes to Opus 4.8 whenever a message is flagged as
"cybersecurity." I'm doing legitimate, defensive security work: building a
personal zero-knowledge password manager (Chrome extension) and fixing a real
user-enumeration vulnerability in my own code. This is exactly the kind of
beneficial, defensive work the stronger model should be allowed to handle.

Three problems:

1. The flag is topic-based, not intent-based. It can't distinguish "harden my

own password manager" from genuinely malicious requests, so it over-triggers
on clearly safe content (your own notification admits this).

1. It silently overrides my explicit "/model Fable 5" choice. A deliberate model

selection shouldn't be reversed without a clear, per-message indication and an
opt-out.

1. For an inherently security-focused project, I get the weaker model precisely

where I need the stronger one most — the whole project keeps tripping the wire.

Request: make the classifier consider intent/context (defensive vs. offensive),
and/or let users explicitly opt to keep their chosen model for flagged-but-
legitimate work. At minimum, make the switch visible and reversible per session.

Environment Info

• Platform: darwin
• Terminal: iTerm.app
• Version: 2.1.170
• Feedback ID: 58dfc209-4dfb-453b-ade4-25e89651d03b

Errors

[]