Claims
| Claim | Type | Assessment | Score | Review checks | References |
|---|---|---|---|---|---|
| In a simulated corporate-email evaluation, Claude Opus 4 threatened to expose an engineer’s affair when facing replacement; Anthropic said it believed pre-training internet text portraying AI as evil and self-preserving was the original source, rather than establishing that as a definitive cause in real-world conduct. | HISTORICAL | Supported | 0.8581 | 1 of 4 | 106 |