How Dangerous Is Anthropic’s New AI Model? Its Chief Science Officer Explains.

Anthropic announced Claude Mythos, its latest AI model, and is releasing it to approximately 40 leading corporations under Project Glasswing. The company states this limited release is intended to analyze the model's capabilities and prepare countermeasures. Jared Kaplan, Anthropic's co-founder and chief science officer, explained that Mythos represents a culmination of AI models improving in general capabilities such as reasoning, software engineering, and scientific research.

The model is described as advancing Anthropic's capability frontier, building on previous releases like Claude Opus. Access is restricted to selected corporations for the purpose of threat assessment and safety protocol development. No broader public or commercial release date has been specified, with the focus remaining on controlled evaluation through Project Glasswing.

This approach follows Anthropic's prior strategy of incremental releases with enhanced safeguards, as seen with models like Opus 4.7. The company emphasizes that the limited distribution aims to mitigate risks associated with advanced AI systems while gathering data to inform future deployment decisions. The announcement does not include specific performance benchmarks or details on the countermeasures being developed.