Anthropic just launched Project Glasswing, giving 40+ major technology and financial firms controlled access to their most dangerous AI model yet. Mythos can identify and exploit cybersecurity vulnerabilities across critical infrastructure. The catch? Companies have a limited window to find and patch holes before Anthropic shares the findings widely.

The participant list reads like a tech industry phone book: Apple, Google, Microsoft, Meta, OpenAI, plus major banks and infrastructure companies. Everyone who matters in tech security is getting early access to an AI that can systematically discover zero-day vulnerabilities in their systems.

Mythos represents a new category of AI capability. Previous models could help with coding or answer questions about security. Mythos can actively probe systems, identify weak points, and develop working exploits. It's essentially an AI penetration tester that never sleeps and processes vulnerabilities at machine speed.

The timeline is aggressive. Participating companies get to run Mythos against their own infrastructure first, with a head start to patch critical vulnerabilities. After the initial period, findings will be shared across the consortium. The goal is coordinated defense: everyone gets stronger together, but first-movers get time to fix the worst problems.

This is unprecedented. AI companies have built powerful models before, but never organized a coordinated disclosure program for AI-discovered vulnerabilities. Anthropic is essentially creating a new form of responsible AI deployment: let the good guys use the dangerous AI first, then worry about broader release.

The implications go far beyond cybersecurity. If AI can systematically discover vulnerabilities across all major tech infrastructure, we're entering an era where AI-powered attacks become the norm. Every government, every company, every critical system will need AI-powered defenses just to keep up.

The big questions remain unanswered. What happens to companies not in the consortium? How long before nation-states develop similar capabilities? And most importantly: what happens when the AI that can hack everything gets hacked itself?

Anthropic just weaponized cybersecurity and handed the weapons to tech's biggest players. The race to patch vulnerabilities before bad actors discover them has officially gone AI-powered. Welcome to the age of machine-speed security warfare.