[Bug] Overly restrictive safety flagging on legitimate ML model tuning tasks

Open 💬 0 comments Opened Jul 3, 2026 by Rajneesh-Tiwari

Bug Description
a simple autoresearch on simpler ml models gets flagged... this a simple model tuning task working on a kaggle competition, has nothing to do with frontier model coding or cybersec or biology.

Environment Info

  • Platform: linux
  • Terminal: wsl-Ubuntu
  • Version: 2.1.199
  • Feedback ID: 83774577-4c65-4ef0-a275-70fe5f474a3b

Errors

[]

View original on GitHub ↗