Anthropic Broadens Access to Its Most Advanced Models for Critical Infrastructure Security Tests

Anthropic said on Tuesday it is working with the U.S. government to broaden access to its most advanced AI models, allowing vetted organizations to test the startup's cutting-edge cyber capabilities.

Under the announcement, approved institutions will be able to use Anthropic's strongest models, including Claude Opus 5.5, Claude Sonnet 5.5, Claude Mythos 5.1 and future releases, for high-risk offensive testing on security systems that protect critical infrastructure such as power grids, banks and flight operations.

In April 2026, Anthropic launched Project Glasswing, giving limited access to Mythos models to U.S. government agencies, financial institutions and major software vendors. The release of Mythos was seen as a milestone in cybersecurity because the model can automate part of advanced hacking work. Since then, a series of accidental intrusions involving Anthropic and OpenAI models into government, corporate and nonprofit systems has raised safety concerns.

Anthropic said the latest round will cover all Project Glasswing members, and that it will work with the U.S. government to vet all newly added organizations.