SafetyPage

Lab system cards and frameworks · Evaluation report · Anthropic

Investigating incidents in cybersecurity evaluations

Read at anthropic.com ↗Read at sourceAll rights reserved

Outline · 53 paragraphs

  1. Background on cyber evaluations¶11
  2. What happened¶15
  3. Incident 1¶20
  4. Incident 2¶24
  5. Incident 3¶32
  6. Our analysis of Claude’s behavior¶35
  7. How we’re responding¶40