Anthropic's Hidden Model: 0% Attack Starts in Tests

Anthropic's internal model had 0% attack starts in replay tests. Compare Mythos 5.1, alignment results, and test limits.