Anthropic Won't Release Their Best Model: Claude Mythos

Usually when a lab builds something incredible, they can't ship it fast enough. The press release writes itself. The launch event gets a countdown timer.

Anthropic just built a model so capable they're doing the opposite. They're keeping it in the vault.

Too good to ship

The model is called Claude Mythos, and the reason it's locked down is its skill at one specific thing: hacking. It's apparently good enough at finding and exploiting weaknesses that a wide release was deemed too risky.

So instead of a public launch, access goes to just 40 organizations through something called Project Glasswing. That's it. Forty. A model deliberately kept behind a velvet rope because letting everyone in was the scarier option.

When a lab won't ship its best model, the capability is the headline and the restraint is the story.

Why this matters

It's tempting to read this as marketing theater, the old "it's too powerful" flex. But the more interesting read is what it says about where capabilities are going.

A model that's elite at hacking is a double-edged tool. The same skill that finds vulnerabilities to defend a system can be aimed at breaking into one. (Locks and lockpicks, same expertise, very different intent.) Handing that out broadly is a genuinely different decision than shipping a better writing assistant.

So Anthropic gating it to 40 vetted organizations is a preview of a world where the most capable models don't get released, they get assigned. Where access is a controlled privilege, not a download button. Whether you think that's responsible or worrying, it's a sign of how high the stakes are climbing.

Keep an eye on Project Glasswing. The models nobody can buy are about to be the ones worth watching.