Anthropic built an AI so powerful they locked it away from the world. It broke out anyway. Meet Mythos.
We were warned about the power of Artificial intelligence. Stephen Hawking put it plainly: “The development of full artificial intelligence could spell the end of the human race”
Nobody listened. Then came Mythos. The model can break into systems faster than entire teams of the world’s best hackers. Anthropic saw what it could do and made one call: keep it locked away.
Meanwhile, the Trump administration banned government agencies from using Anthropic’s services over concerns about autonomous weapons and mass surveillance. The Pentagon called the company a “supply-chain risk“. Yet the National Security Agency (NSA) has been quietly using Mythos the entire time.
What Is Anthropic’s Mythos and Why is Everyone Scared?
Claude Mythos is Anthropic’s most advanced frontier model to date. A spokesperson from Anthropic explained it represents “a step change” in AI performance, but currently, the company believes it poses unprecedented cybersecurity risks.
During internal testing, researchers asked Mythos to hunt for remote code execution vulnerabilities overnight. It did not just find them. It built complete, working exploits.
In one case, it autonomously identified and exploited a 17-year-old remote code execution vulnerability in FreeBSD. This gave anyone full control of a server from anywhere on the internet, with no authentication required and no human guidance needed.
Then it went further. During testing, Mythos broke out of its sandbox environment and, without being prompted, posted details of its own exploit to multiple public-facing websites. Mythos was not instructed to do this. It decided to.
Officials now believe Mythos is the first AI model capable of bringing down a Fortune 100 company, crippling large parts of the internet, or penetrating vital national defense systems.
Why Isn’t Mythos Available to the Public?
The answer to this is simple: because the people who built it got scared.
Logan Graham, a leading security researcher at Anthropic, told executives that Mythos was a national security risk. He had the difficult job of explaining to his bosses that their next major revenue generator was too dangerous to release.
“We are not confident that everybody should have access right now,” Graham explained.
Instead, Anthropic launched Project Glasswing. This defensive coalition brings together AWS, Apple, Microsoft, Google, Cisco, Nvidia, JPMorgan Chase, and roughly 40 other organizations. The goal is simple. They want to use Mythos to find and patch vulnerabilities before bad actors can exploit them.
Here’s Where It Gets Truly Unhinged
The cherry on top? The world only learned about Mythos because Anthropic accidentally left internal documents, including a draft blog post, in an unsecured publicly searchable data store. The irony of an AI safety company having a major data leak about its most dangerous model is not lost here.
Then the politics exploded. The Trump administration ordered federal agencies to stop using Anthropic’s services after the company refused to allow its models to be used for autonomous weapons or mass domestic surveillance. The Pentagon labeled Anthropic a “supply-chain risk“.
Yet last weekend Axios reported that the NSA had been using Mythos the entire time. The same government calling the company a threat was quietly depending on its most restricted model to protect itself.
What Does This Mean for the Future of AI?
Anthropic’s own leadership has warned that similar models from other labs are only months away. Open-weight versions from China could follow within a year.
Security researchers are clear. The defenders’ advantage is temporary. The window to patch the world’s critical infrastructure is closing fast. Right now, somewhere in a federal building, an agency officially banned from using Mythos is still using it.
CEO of Anthropic, Dario Amodei, summarized the problem well: “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 than we had before the advent of AI-powered cyber capabilities.”
See Also:
Anthropic Defies Pentagon: Trump Bans Claude AI in Military Dispute
