Anthropic — AI Safety Level: ASL-3 (Opus 4, Opus 4.5, Opus 4.6), ASL-2 (Sonnet, Haiku)
Reset by flagship-curate before re-verification
Our claim
entire record- Subject
- Anthropic
- Property
- AI Safety Level
- Value
- ASL-3 (Opus 4, Opus 4.5, Opus 4.6), ASL-2 (Sonnet, Haiku)
- As Of
- February 2026
- Notes
- Opus 4.6 launched Feb 2026 under ASL-3; all Opus-tier models are ASL-3, Sonnet/Haiku remain ASL-2
Source evidence
1 src · 2 checksNoteThe source confirms Claude Opus 4 is ASL-3 and states Claude Sonnet 4 does not need ASL-3 (implying it remains ASL-2). However, the claim lists Opus 4.5, Opus 4.6, Sonnet, and Haiku with specific ASL levels, which are not addressed in the source text. The source only discusses Opus 4 and Sonnet 4 explicitly. Without information about the other models mentioned in the claim, the claim cannot be fully verified.
NoteThe source text is dated May 22, 2025, and discusses the activation of ASL-3 for Claude Opus 4 only. The claim references February 2026 information (Opus 4.6 launch, current status of all models as of 2026-02), which is 9 months in the future relative to the source date. The source does not address Opus 4.5, Opus 4.6, or provide information about model classifications as of February 2026. While the source confirms Opus 4 is ASL-3 and implies Sonnet remains ASL-2, it cannot verify the specific claim about future model releases and their safety levels. This is a temporal mismatch issue—the claim makes assertions about a future date that the source cannot address.