Anthropic — AI Safety Level: ASL-3 (Opus 4+), ASL-2 (Sonnet/Haiku)

NoteThe source directly confirms that ASL-3 protections were activated for Claude Opus 4 as of May 22, 2025 (matching the claim's 2025-05 date). The source explicitly states 'we have ruled out that Claude Sonnet 4 needs the ASL-3 Standard' and that all previous models were deployed under ASL-2, which covers Sonnet and Haiku models. The claim's characterization of ASL-3 for Opus 4 and ASL-2 for Sonnet/Haiku is accurate. The source also confirms this is 'precautionary and provisional' as stated in the claim's additional context.

NoteThe source directly confirms the claim's core assertions: (1) Claude Opus 4 is deployed with ASL-3 protections as of May 22, 2025, (2) this activation is explicitly described as 'precautionary and provisional', (3) Sonnet (Sonnet 4 specifically mentioned) does not require ASL-3 and remains at ASL-2 baseline, and (4) Haiku is not mentioned but the source confirms Sonnet does not meet ASL-3 threshold. The date matches the claimed 2025-05 timeframe. The only minor gap is that the source doesn't explicitly mention Haiku's ASL level, but it does confirm Sonnet's ASL-2 status and Opus 4's ASL-3 status with the precautionary/provisional qualifier as claimed.