SafetyPage

Evaluators and safety organisations · Article · Goodfire

Training and Deploying Production Cyber Monitors on Kimi K3

8 Oct 2026

Read at goodfire.com ↗Read at sourceNo licence statement found

Outline · 58 paragraphs

  1. Introduction¶1
  2. Our Monitor Recipe¶12
  3. Data¶13
  4. Architecture: a monitor cascade¶16
  5. Inference Optimizations¶21
  6. Results¶27
  7. Evaluating Our Monitors¶28
  8. Comparing Different Monitoring Systems¶35
  9. External Red Teaming Results¶40
  10. Conclusion¶45
  11. Appendix¶49
  12. Comparing on vs. off policy monitors¶50
  13. Reducing judge calls while preserving recall¶52
  14. Footnotes¶54