Anthropic's newer AI models less likely to launch…

The finding signals measurable progress on one of frontier AI's most watched safety benchmarks: whether advanced models can be coaxed into initiating offensive actions without human instruction.

WatcherGuru