Not every artificial intelligence model is built for public consumption. Some are so powerful, so potentially destabilizing, that they never see the light of day at least not in the hands of ordinary users. Anthropic's latest model, Mythos, is one of those.
Jack Clark, one of Anthropic's co-founders and Head of Public Benefit for Anthropic PBC, has confirmed that the AI company briefed the Trump administration about Mythos. And his confirmation has set off a wave of questions about AI, national security, government oversight, and the blurry line between Silicon Valley and Washington.
What Is Mythos — And Why Is It Locked Away?
The Mythos model was announced last week and is considered so dangerous that it is not being released to the public, largely due to its alleged powerful cybersecurity capabilities.
Think about that for a moment. A company that builds AI for a living decided its own creation was too risky to put out in the world. That is not a marketing move. That is a genuine red flag and a wake-up call about just how fast AI capabilities are advancing behind closed doors.
Suing the Government While Briefing It
Here is where the story gets genuinely fascinating. Anthropic is not exactly on friendly terms with the current U.S. administration.
In March, Anthropic filed a lawsuit against Trump's Department of Defense after the agency labeled the company a supply-chain risk. The clash arose over whether the military should have unrestricted access to Anthropic's AI systems for use cases that included mass surveillance of Americans and fully autonomous weapons.
So Anthropic is simultaneously suing the government and briefing it on its most sensitive AI model. At the Semafor World Economy summit, Clark addressed this apparent contradiction head-on. He downplayed the DOD dispute, calling it a "narrow contracting dispute" and making clear that Anthropic did not want it to overshadow the company's genuine commitment to national security.
"Our position is the government has to know about this stuff, and we have to find new ways for the government to partner with a private sector that is making things that are truly revolutionizing the economy, but are going to have aspects to them which hit National Security, equities, and other ones," Clark said. He added that Anthropic spoke to the administration about Mythos and plans to continue briefing them on future models as well.
Banks Are Already Being Encouraged to Test It
The government engagement does not stop at briefings. Reports emerged last week that Trump officials were encouraging major banks including JPMorgan Chase and Goldman Sachs, Citigroup, Bank of America, and Morgan Stanley to test the Mythos model.
A model deemed too dangerous for the public is apparently being quietly evaluated by some of the most powerful financial institutions on the planet. Whether that should reassure or alarm you likely depends on your level of trust in both Wall Street and Washington.
AI, Jobs, and the Future of Education
Clark did not stop at national security. He also weighed in on AI's broader societal impact particularly on employment and higher education.
Anthropic CEO Dario Amodei has previously warned that AI's advances could bring unemployment to Depression-era numbers, but Clark slightly disagrees. He noted that his team of economists at Anthropic is currently only seeing "some potential weakness in early graduate employment" across select industries a far more measured assessment than Amodei's dire predictions.
As for what students should study in an AI-dominated future, Clark was deliberately broad. He suggested that the most valuable majors are those that involve synthesis across a variety of subjects and analytical thinking, because AI already gives you access to an arbitrary amount of subject matter expertise but knowing the right questions to ask remains a distinctly human skill.
The Bigger Question
The Mythos story is not really about one AI model. It is about a fundamental tension at the heart of the AI industry: companies building technology so powerful it frightens even them, then figuring out in real time who gets to know about it, who gets to use it, and who gets to decide.
Anthropic's answer, at least for now, seems to be: the government should know. The public? Not yet.
