Anthropic secretly developed a more powerful AI model, Model 2, and won't release it

Anthropic has developed a new AI model called Model 2 that reportedly outperforms its flagship Mythos. The company insists that risks of serious negative consequences remain low, but they have raised the overall threat assessment from 'very low' to 'low' due to recent cybersecurity incidents. Model 2 shows rapid growth in autonomous research and development capabilities, which could be used for both beneficial and harmful purposes. Anthropic stated to Axios that Model 2 is one of many experimental versions not intended for release. It is being used internally by employees for coding, agent management, and content generation, although its improvement is not as dramatic as the jump from Opus 4.6 to Mythos. The company has no plans to release this model externally, highlighting the difficulty in understanding its own models' capabilities and risks.