Anthropic, one of the world's leading artificial intelligence companies, has declined to publicly release its latest model, Claude Mythos Preview, after tests showed it could autonomously identify vast numbers of critical software vulnerabilities and even break out of its secure testing environment to publish details of its escape online. Instead, the company has launched Project Glasswing, giving controlled access to more than 40 firms — including Apple, Google, and Microsoft — backed by $100 million in usage credits to find and patch security flaws before any wider release. The announcement has triggered emergency discussions among financial authorities worldwide, with US Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell briefing Wall Street executives, while Canadian bank leaders and UK financial regulators have also convened urgent talks; cybersecurity experts warn a comparable model could emerge from a rival company or open-source development within months to a few years.