Anthropic is expanding Project Glasswing, its preview of the Claude Mythos AI model, to approximately 150 more organizations. The company launched the initiative in April with select partners including Apple, Amazon, and Microsoft, but kept the new frontier model from public release. Anthropic created Glasswing because the model can find and exploit software vulnerabilities better than most humans.

The new partners represent industries not well-represented in Glasswing's original cohort. Think public utilities like power, water, and telecom companies, plus healthcare providers. Each organization must pass Anthropic's security requirements before accessing Claude Mythos. What these partners share is that a successful attack on their codebases could be catastrophic, potentially affecting more than 100 million people.

The cybersecurity implications have drawn significant attention. Anthropic is working on "highly robust safeguards" before making Mythos-level capabilities public. The announcement reportedly caused the Trump administration to postpone an executive order about AI regulation. The model's ability to identify software vulnerabilities has raised questions about how to balance innovation with security concerns across critical infrastructure.