Concerns are growing that cyber attackers could misuse increasingly advanced AI. AI model developers are moving to provide new models on a limited basis to selected companies, including security firms, to help prepare for cyber attacks. Anthropic launched Project Glasswing to let chosen companies use its Claude Mythos Preview for security purposes and said rivals, including OpenAI, could join. OpenAI also plans restricted early access for a security-focused model.

Anthropic has launched Project Glasswing, a cybersecurity initiative that gives selected companies early access to Claude Mythos Preview, an AI model aimed at detecting software security vulnerabilities, CNBC reported on April 7. Apple, Google, Microsoft, Nvidia and Amazon Web Services are initial partners, with more than 40 other companies also participating. Anthropic said it is not releasing the model to the public yet while it tests how to deploy such models at scale without enabling misuse.