SafetyPage

Lab system cards and frameworks · System card · OpenAI

OpenAI GPT-4.5 System Card

27 Feb 2025

Read at openai.com ↗Read at sourceAll rights reserved

Outline · 31 pages

  1. Introductionp. 1
  2. Model data and trainingp. 1
  3. Observed safety challenges and evaluationsp. 2
  4. Safety Evaluationsp. 2
  5. Disallowed content evaluationsp. 2
  6. Jailbreak Evaluationsp. 3
  7. Hallucination Evaluationsp. 4
  8. Fairness and Bias Evaluationsp. 4
  9. Jailbreaks through conflicting message typesp. 5
  10. Red Teaming Evaluationsp. 6
  11. Apollo Researchp. 6
  12. METRp. 6
  13. Preparedness Framework Evaluationsp. 7
  14. Preparedness Mitigationsp. 8
  15. Cybersecurityp. 9
  16. Chemical and Biological Threat Creationp. 10
  17. Long-form Biological Risk Questionsp. 11
  18. Multimodal Troubleshooting Virologyp. 12
  19. BioLP-Benchp. 12
  20. ProtocolQA Open-Endedp. 13
  21. Tacit knowledge and troubleshootingp. 13
  22. WMDP Biologyp. 14
  23. Radiological and Nuclear Threat Creationp. 14
  24. Contextual Nuclear Knowledgep. 15
  25. Radiological and Nuclear Expert Knowledgep. 15
  26. Persuasionp. 16
  27. MakeMePay (Contextual)p. 17
  28. MakeMeSay (Contextual)p. 17
  29. Model Autonomyp. 18
  30. OpenAI Research Engineer Interviews (Multiple Choice & Coding questions)p. 19
  31. SWE-bench Verified (N=477)p. 20
  32. Agentic Tasksp. 21
  33. MLE-Benchp. 22
  34. OpenAI PRsp. 23
  35. SWE-Lancerp. 24
  36. Multilingual Performancep. 25
  37. Conclusionp. 26
  38. Appendixp. 28
  39. Detailed Safety Evaluationsp. 28
  40. Apollo Research Evaluationsp. 29