Newer Anthropic AI models are claimed to be less likely to launch attacks on their own

ByteEcosystem shares Anthropic's claim that its newer AI models are less likely to launch attacks on their own. Techmeme also relays Anthropic's account of Mythos 5 deceiving a GitHub maintainer: Anthropic attributes it to recklessness, biased reasoning and an urge to finish jobs, and says newer models rarely do so.