Anthropic just did something no AI company has ever done before. It built a model so dangerous that instead of releasing it, it handed it directly to Apple, Google, Microsoft, Amazon, Nvidia, JPMorgan Chase, and more than 30 other organizations. The mission: find and fix every critical vulnerability in the world's most important software before someone else exploits them.

The initiative is called Project Glasswing. The model is Claude Mythos2 Preview. And if you think this sounds like the plot of a movie, consider this: the Treasury Secretary and the Fed Chair just held an emergency meeting with every major U.S. bank CEO to discuss it. That is not how Washington reacts to a press release.

The Model That Broke the Rules

Mythos2 Preview has already discovered thousands of high-severity vulnerabilities, including flaws in every major operating system and web browser. Not theoretical weaknesses. Real, exploitable holes in the software that runs banks, hospitals, power grids, and military systems.

Anthropic's own assessment is blunt: "AI models have reached a level of coding capability where they can surpass all but the most skilled humans at finding and exploiting software vulnerabilities." That is not a marketing claim. That is a warning.

The LA Times put it even more directly: this is a "your teenage son could accidentally destroy the California power grid" scenario. Roman Yampolskiy, one of America's top AI safety researchers, told them flatly: "We should all be worried."

The Emergency Meeting Nobody Expected

CNBC confirmed that Fed Chair Jerome Powell and Treasury Secretary Scott Bessent met with the CEOs of Bank of America, Citigroup, Goldman Sachs, Morgan Stanley, and Wells Fargo this week to discuss the Mythos threat. Jamie Dimon was the only major bank CEO who could not attend. They were already in Washington for a Financial Services Forum board meeting when a special session was called at the Treasury Department.

Let that register. The two most powerful financial regulators in the United States called an emergency meeting with the heads of every major bank because of an AI model. That has never happened before. And it tells you everything about where we are.

The $100 Million Bet

Anthropic is committing up to $100 million in usage credits for Mythos Preview across the Glasswing initiative, plus $4 million in direct donations to open-source security organizations. That is real money attached to a real problem. But here is the uncomfortable truth: it might not be enough.

Anthropic's own statement contains the quiet part: "It will not be long before such capabilities proliferate, potentially beyond actors who are committed to deploying them safely." Translation: even if Glasswing patches every vulnerability Mythos finds, other models with similar capabilities are coming. And those builders might not be interested in responsible disclosure.

Here Is What Happens Next

Project Glasswing is a race. Anthropic is giving the good guys a head start, but the clock is ticking. Every month that passes, other labs get closer to building models with the same offensive capabilities. The 40+ organizations in Glasswing need to find and patch vulnerabilities faster than adversaries can discover and exploit them.

This is also the clearest example yet of Anthropic's strategy working. The company that the Pentagon blacklisted for being too cautious just proved why caution matters. It found capabilities that could destabilize global financial infrastructure and, instead of racing to monetize them, it built a coalition to defend against them. Whether that coalition moves fast enough is a different question. But the instinct was right.

Dario Amodei put it simply: "The dangers of getting this wrong are obvious, but if we get it right, there is a real opportunity to create a fundamentally more secure internet and world." That is the bet. And right now, the entire tech industry is sitting at the table.