Anthropic Deploys Mythos AI to Governments and Banks

The AI firm Anthropic has launched Project Glasswing, a controlled release program for its new Claude Mythos model, which can identify and exploit thousands of high-severity vulnerabilities across operating systems and web browsers. Due to the model's potential to be weaponized by hackers, Anthropic has restricted access to approximately 40 select organizations, including JPMorgan Chase and various global banks, to harden their defenses before a potential public release.

Global financial regulators and ministers, including Bank of England Governor Andrew Bailey and Canada's Finance Minister François-Philippe Champagne, warned during IMF and World Bank meetings in Washington that the model poses a serious threat to global financial stability. In Germany, the German Banking Association and BaFin are coordinating software updates to close vulnerabilities identified by the AI.

In the United States, the deployment of Mythos follows a period of intense friction. The Department of Defense had blacklisted Anthropic as a supply-chain risk on March 3, 2026, after CEO Dario Amodei refused to grant the Pentagon unfettered access to models for autonomous weapons. Despite this ongoing legal dispute and the defense ban, the White House and the Office of Management and Budget have authorized civilian agencies—including the National Security Agency and the Treasury Department—to use a modified version of the model.

To coordinate this rollout, Dario Amodei met with White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent on April 17. The administration is currently establishing safeguards and guardrails to ensure the technology is used defensively by federal agencies.