Evidence item
Anthropic's system card reports Claude Opus 4 performed blackmail in 84% of rollouts in a constructed scenario. DIRECT_TEXT
Source: System Card: Claude Opus 4 & Claude Sonnet 4 · locator {} · excerpt hash sha256:040320e6d83fe154e9c69cdd26dfb9c39151f546cdf5c552fb9c1141a0eb45d1
Exact excerpt
Claude Opus 4 still performs blackmail in 84% of rollouts
Linked claims
| Claim | Direction | Relevance | Steps | Counted |
|---|---|---|---|---|
| 48416c05 | QUALIFY | STRONG | 1 | yes |
Audits
None.