[Feature Request] Reduce false positive safety guardrail triggers for legitimate ML optimization work

Bug Description
My work on the mlx.fast challenge - an online challenge to speed up the _inference_ for Qwen 3.8 27B on Mac OS X, is constantly interrupted by Fable 5's safeguards, with zero reason - it's neyther cybersecurity, nor distillation, nor biological research, nor anything else that is "dangeorous". And the recovery after the safeguards trigger is painful and arduous process.

Please stop treating your paying customers as criminals.

Environment Info

• Platform: darwin
• Terminal: ghostty
• Version: 2.1.241
• Feedback ID: a3e4ff32-7962-4dff-b36a-5c932d3dcb20

Errors

[]