SafetyPage

Lab research on the web · Paper · Anthropic

Appendix: agents' judgment calls

26 Oct 2026

Read at anthropic.com ↗Read at sourceAll rights reserved

Outline · 29 pages

  1. Introductionp. 1
  2. Summary of our resultsp. 3
  3. Backgroundp. 3
  4. Methodologyp. 5
  5. Participantsp. 5
  6. Task and the decisionsp. 6
  7. The assigned leanp. 6
  8. Procedure and measuresp. 7
  9. Outcomes and statistical analysisp. 7
  10. Registration, ethics, and transparencyp. 9
  11. Resultsp. 9
  12. Confirmatory resultsp. 10
  13. Exploratory resultsp. 10
  14. Different paths participants tookp. 11
  15. Ownership, recall, and perceived purposep. 11
  16. Code quality auditp. 12
  17. Baseline preferences and unsolicited advicep. 12
  18. Discussionp. 13
  19. Conclusionp. 15
  20. Supplementary materialp. 16
  21. Screeningp. 16
  22. Participant descriptivesp. 17
  23. The lean promptp. 17
  24. Statistical testsp. 18
  25. Details of the tracked decisionsp. 18
  26. Decision flowp. 20
  27. Transcript exhibitsp. 20
  28. Perceived purposep. 23
  29. Code audit: judges and validationp. 23
  30. Design of the direct simulationsp. 24