Anthropic Opens Mythos-Class AI to Public With Safety Guardrails

Anthropic has launched Claude Duet, an app representing the company's push into AI agents. The app is currently available through a waitlist limited to Pro and Max plan users. This move signals Anthropic's strategic focus on agentic AI capabilities as a key product direction.

A coalition of AI safety advocates is proposing that OpenAI, Anthropic, and other AI labs cap the computational resources allocated to research and development. The proposal would require companies to dedicate minimum shares of compute to safety monitoring and customer inference, leaving less capacity for training new models. The idea emerges as industry leaders like Sam Altman explore various governance models to ensure safeguards keep pace with AI capability advances.

Anthropic has proposed an AI safety bill that would require federal agencies to obtain judicial approval before blocking an AI model's release, even if the model poses risks. The proposal appears designed to protect against regulatory overreach while establishing a high threshold for government intervention. The mechanism creates a legal requirement for court involvement in model blocking decisions, potentially limiting agency authority in AI safety matters.

Nscale and Anthropic are in reported talks for a $45 billion compute deal, according to sources. If completed, the agreement would represent the latest in a series of major compute infrastructure agreements Anthropic has negotiated over the past year. The deal underscores the massive capital requirements driving competition in large language model development.