OpenAI — AI Safety Level: High/Critical capability thresholds (Preparedness Framework v2)

NoteThe claim accurately describes OpenAI's Preparedness Framework v2 (April 2025). The source confirms: (1) OpenAI uses its own Preparedness Framework rather than Anthropic's ASL system; (2) Version 2 was last updated 15 April 2025, matching the 'as of 2025-04' date in the claim; (3) Two capability thresholds are defined: 'High capability' and 'Critical capability' with the descriptions provided matching the claim; (4) The framework applies to three Tracked Categories (Biological/Chemical, Cybersecurity, AI Self-improvement). The claim's statement that 'No models currently rated at Critical level' is consistent with the source's language about preparing safeguards for future models and the conditional nature of Critical threshold discussions, though the source does not explicitly state this in a single sentence.