White House Grants US Agencies Access to Anthropic Mythos: Bloomberg

• Mythos was announced April 7 and has found 'thousands' of major vulnerabilities in operating systems, web browsers, and other software through its high-level coding abilities
• Federal CIO Gregory Barbaccia confirmed the government is coordinating with model providers, industry partners, and intelligence community to establish guardrails before agency deployment
• The initiative continues despite Pentagon cutting business ties with Anthropic following a contract dispute, though the company remains in dialogue with the Trump administration

Summary

The U.S. government plans to grant major federal agencies access to Anthropic's advanced AI model, Mythos, despite significant cybersecurity concerns, according to Bloomberg News. The deployment is part of "Project Glasswing," a controlled initiative announced April 7 that permits select organizations to use the unreleased Claude Mythos Preview model exclusively for defensive cybersecurity purposes.

Key Capabilities and Concerns:

Mythos has identified "thousands" of major vulnerabilities across operating systems, web browsers, and other software. The model's advanced coding capabilities give it unprecedented ability to discover and potentially exploit security weaknesses, raising alarm among cybersecurity experts.

Government Response:

Gregory Barbaccia, federal chief information officer at the White House Office of Management and Budget, confirmed the administration is collaborating with model providers, industry partners, and the intelligence community to establish appropriate guardrails before releasing a modified version to agencies.

Company Context:

Anthropic co-founder Jack Clark stated Monday that the AI lab is working with the Trump administration, despite the Pentagon recently severing business ties with the company following a contract dispute.

Market Implications:

This development highlights the dual-use nature of frontier AI models and growing government involvement in AI deployment. The initiative underscores tensions between leveraging advanced AI for defensive cybersecurity versus risks of the technology falling into wrong hands or being exploited maliciously.

Neither the White House nor Anthropic provided immediate comment to Reuters' requests. This marks a significant step in government adoption of cutting-edge AI tools amid ongoing debates about AI safety and regulation.