Anthropic just made a big move. On Tuesday, the AI company released an early preview of its new frontier model called Mythos. And it is not for everyone. Only a select group of partner companies will get to use it and only for cybersecurity work.
What Is Mythos?
Mythos is a general-purpose AI model built for Anthropic's Claude systems. Think of it as a much bigger, much smarter version of what already exists. The company says it has strong coding and reasoning abilities, especially when it comes to working like an agent.
Before this official preview, Mythos was accidentally leaked last month through a security slip-up. A draft blog post about the model was found sitting in an unsecured public data lake. At that time, the model had a codename: Capybara. The leaked document described it as larger and more intelligent than Anthropic's Opus models, calling it the most powerful AI the company had ever built.
Now it is here. Sort of.
Project Glasswing: A Cybersecurity-First Launch
Anthropic is not releasing Mythos to the public. Instead, it is rolling it out through a new initiative called Project Glasswing.
More than 40 partner organizations are part of this project. The goal is simple: use Mythos for defensive security work and to find vulnerabilities in critical software. The model was not specifically trained for cybersecurity, but Anthropic says it is already very good at it.
How good? Over the past few weeks, Mythos has reportedly found thousands of zero-day vulnerabilities. Many of them are critical. Some are one to two decades old. That is a serious find.
Big Names Are Already On Board
The list of partners reads like a tech hall of fame. Amazon, Apple, Broadcom, Cisco, CrowdStrike, the Linux Foundation, Microsoft, and Palo Alto Networks are all part of the initiative.
These companies will use the model, learn from it, and then share their findings with the broader tech industry. That is the deal. Everyone benefits, at least in theory.
The Government Angle Is Complicated
Anthropic says it has been in "ongoing discussions" with federal officials about Mythos. But here is where things get messy.
Anthropic and the Trump administration are currently in a legal fight. The Pentagon labeled the AI lab a supply-chain risk after Anthropic refused to allow its AI to be used for autonomous targeting or surveillance of American citizens. So any government collaboration right now is walking on thin ice.
A Rough Month for Anthropic
This launch comes during what has been a turbulent stretch for the company.
Last month, Anthropic accidentally exposed nearly 2,000 source code files and over half a million lines of code. The leak happened because of a mistake during the launch of Claude Code version 2.1.88.
It got worse. While trying to clean up that mess, Anthropic accidentally caused thousands of GitHub repositories to be taken down. The company called it an accident. The internet called it a bad week.
And then there was the original Mythos leak itself. The company blamed "human error" for the draft blog ending up in a publicly accessible location.
Three incidents in one month. Not a great look for a company launching a cybersecurity product.
Why This Matters
Here is the thing. AI models that can find zero-day vulnerabilities are a double-edged sword. Anthropic itself acknowledged in the leaked documents that Mythos could pose a cybersecurity threat if bad actors used it to find and exploit bugs instead of fixing them.
That is exactly why Project Glasswing keeps the model locked down. Only trusted partners get access. No public release.
Still, the potential is massive. Decades-old bugs sitting in critical software, found in weeks by an AI model. That could change how the entire industry approaches security.
The Bottom Line
Anthropic is betting big on Mythos. The model is powerful. The partners are impressive. The cybersecurity use case is smart.
But the timing is rough. Between source code leaks, GitHub takedowns, and a legal battle with the government, Anthropic has a lot of trust to rebuild. Launching a security initiative while dealing with your own security problems is a bold choice.
Whether Mythos lives up to the hype will depend on what those 40+ partners actually find and whether Anthropic can keep its own house in order while it helps secure everyone else's.
