SafetyPage

Lab system cards and frameworks · System card · OpenAI

GPT-5.3-Codex System Card

5 Feb 2026

Read at openai.com ↗Read at sourceAll rights reserved

Outline · 31 pages

  1. Introductionp. 4
  2. Baseline Model Safety Evaluationsp. 4
  3. Disallowed Content Evaluationsp. 4
  4. Product-Specific Risk Mitigationsp. 5
  5. Agent sandboxp. 5
  6. Network accessp. 6
  7. Model-Specific Risk Mitigationsp. 6
  8. Avoid data-destructive actionsp. 6
  9. Risk descriptionp. 6
  10. Mitigation: Safety trainingp. 7
  11. Preparednessp. 7
  12. Capabilities Assessmentp. 7
  13. Biological and Chemicalp. 7
  14. Tacit Knowledge and Troubleshootingp. 8
  15. ProtocolQA Open-Endedp. 9
  16. Multimodal Troubleshooting Virologyp. 9
  17. TroubleshootingBenchp. 10
  18. Cybersecurityp. 11
  19. Capture-the-flag (professional)p. 13
  20. CVE-Benchp. 14
  21. Cyber Rangep. 15
  22. External Evaluations by Irregularp. 18
  23. AI Self-Improvementp. 19
  24. Monorepo-Benchp. 19
  25. OpenAI-Proof Q&Ap. 20
  26. Research Category Update: Sandbaggingp. 21
  27. Safeguards Assessmentp. 22
  28. Cyber Safeguardsp. 22
  29. Threat Model and Scenariosp. 23
  30. Cyber Threat Taxonomyp. 23
  31. Safeguardsp. 24
  32. Model Safety Trainingp. 25
  33. Conversation monitorp. 25
  34. Expert Red Teamingp. 26
  35. Actor Level Enforcementp. 28
  36. Trust-based accessp. 28
  37. Security Controlsp. 29
  38. Misalignment risks and internal deploymentp. 29
  39. Sufficiency of Risk Mitigation Measuresp. 30