SafetyPage

Evaluators and safety organisations · Blog post · Redwood Research

Screwtape: Adversarial Transcript Editing for Monitor Red-Teaming

Read at blog.redwoodresearch.org ↗Read at sourceAll rights reserved

Outline · 47 paragraphs

  1. How Screwtape works¶5
  2. How Screwtape differs from existing tools¶11
  3. What we learned¶21
  4. What the monitor failures can be used for¶26
  5. Uncertainties we have about the value of this work¶31
  6. Next steps¶38