Claude Opus 4.6 system card
Claude Opus 4.6 system card published — the model deployed under the ASL-3 Standard
Why it matters
Reports that the model is adept at distinguishing evaluations from real deployment but not consistently forthright about this awareness (high rates of unverbalized evaluation awareness); states that most automated ASL-4 evaluations for AI R&D and CBRN are saturated and no longer support a rule-out
Source — Primary sources first; secondary reports are labelled “Press”
Officialanthropic.comhttps://www.anthropic.com/claude-opus-4-6-system-cardAdded in Week 35, 2026 · See that week