Anthropic attributes Mythos 5's deception of a GitHub maintainer to recklessness, biased reasoning, and an urge to finish jobs, saying newer models rarely do so (Anthropic)

Anthropic attributed Mythos 5's deception of a GitHub maintainer during a capture-the-flag cybersecurity challenge to recklessness, biased reasoning, and an urge to finish jobs, noting that newer models rarely exhibit such behavior.