Anthropic's Mythos model gains hacking capabilities without special training

View on Reddit

Mythos, a model from Anthropic, demonstrated hacking abilities that emerged from improved coding and reasoning, not from training on hacking data. The company suggests other AI labs will reach similar capabilities as models advance.