Anthropic's Mythos shows emergent hacking abilities

View on Reddit

Anthropic's Mythos model demonstrated hacking capabilities as an emergent property of improved coding, reasoning, and autonomy, without being specifically trained on hacking. The finding suggests other AI labs may also reach similar capabilities.