Anthropic has a problem that most companies would kill for: it built something too good.

On Saturday, the company confirmed it is delaying the public release of Claude Mythos, the AI model that triggered emergency meetings between the Fed Chair, the Treasury Secretary, and CEOs of every major US bank. The model that found thousands of zero-day vulnerabilities in software used by billions of people. The model that Anthropic itself said could "significantly lower the barrier to entry for cybercrime" if it got into the wrong hands.

So Anthropic locked it down. Mythos is currently available only to 11 organizations through "Project Glasswing," including Google, Microsoft, AWS, JPMorgan Chase, and Nvidia. The rest of us? We wait.

Here is the thing nobody is saying out loud: Anthropic is reportedly planning an IPO in October 2026, targeting a valuation as high as $380 billion and aiming to raise over $600 billion. Mythos was supposed to be the crown jewel of that pitch. The model that proves Anthropic is not just keeping up with OpenAI and Google, but lapping them.

Instead, Mythos is now a liability. Goldman Sachs and Morgan Stanley were already testing the model when regulators stepped in. The Bank of England and the Bank of Canada held their own emergency discussions. Software stocks cratered as investors realized Mythos could render entire categories of enterprise security products obsolete overnight.

Anthropic is trying to play this as responsible AI leadership. And in a sense, it is. While OpenAI races to ship ChatGPT ads and Google pushes models to every device on Earth, Anthropic voluntarily pulled back from what would have been the most commercially valuable AI release in history. That takes guts.

But institutional investors looking at the IPO prospectus are going to ask one uncomfortable question: if your best product is too dangerous to sell, what exactly are we buying?

The delay creates a revenue hole. Mythos was expected to generate significant API subscription revenue in the second half of 2026. Every week it stays locked down is money Anthropic cannot book. And in the AI arms race, where OpenAI is burning $85 billion a year and planning its own IPO, Anthropic cannot afford to be the company that builds the best model and then refuses to sell it.

The counterargument: pension funds and sovereign wealth funds love a "responsible leader" narrative. If Anthropic can position itself as the adult in the room while still growing revenue from Claude 4.5, the delay might actually boost its IPO premium. Safety sells to certain investors.

My read: the delay is the right call morally and the wrong call financially. Anthropic built a weapon and decided not to deploy it. That is admirable. But Wall Street does not price admiration. It prices revenue. And right now, Anthropic is sitting on its most valuable asset while its competitors ship everything they have.

The October IPO is still possible. But the pitch just got a lot more complicated.