SafetyPage

Lab system cards and frameworks · Framework · Anthropic

Frontier Safety Roadmap

19 Oct 2026

Read at anthropic.com ↗Read at sourceAll rights reserved

Outline · 61 paragraphs

  1. Anthropic's Frontier Safety Roadmap¶1
  2. Security¶3
  3. Safeguards¶5
  4. Alignment¶7
  5. Policy¶9
  6. Updates¶17
  7. Leveling up across the board¶31
  8. Fully automated attack investigations¶33
  9. Moving toward an “eyes on everything” state for our internal AI development activities¶35
  10. Policy proposal on embedded third-party evaluators¶37
  11. Policy framework for targeting the pace of AI development¶40
  12. Update on “Upholding Claude’s Constitution”¶43
  13. Today’s capabilities and mitigations February 22nd, 2026¶45
  14. Potential advances in the next few months¶50
  15. Automated R&D¶53