Anthropic's Mythos shows emergent hacking capabilities

View on Reddit

Mythos, a new Anthropic model, exhibits hacking abilities as an emergent property of improved coding, reasoning, and autonomy, not from specific training on hacking. Anthropic notes that other AI labs will likely reach similar capabilities in the future.