Article By Frank Bergman
Federal officials are scrambling after a powerful new artificial intelligence (AI) model demonstrated the ability to hack virtually every major operating system and web browser, triggering urgent warnings from top government and financial leaders.
AI giant Anthropic’s new system, known as “Mythos,” is being kept under tight restrictions.
However, insiders say the threat is already serious enough that the U.S. government is racing to understand it before it’s too late.
Treasury Rushes to Access High-Risk AI
According to reports, the U.S. Treasury Department is urgently seeking access to Anthropic’s restricted model
Treasury Chief Information Officer Sam Corcos has already briefed cybersecurity teams and is pushing to get access to Mythos “as soon as this week,” according to a source cited by Bloomberg.
The move comes after Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell held an emergency meeting with Wall Street executives.
They are warning that this new generation of AI could unleash a wave of cyberattacks on major financial systems, unlike anything seen before.
Bank leaders were reportedly told to take the threat seriously and even begin using the system themselves to hunt for vulnerabilities before bad actors do.
AI That Can Break Into Everything
The concern is not theoretical.
Anthropic’s own internal testing revealed that Mythos can identify and exploit vulnerabilities “in every major operating system and every major web browser when directed by a user to do so.”
In one documented case, the AI:
• Built a working browser exploit
• Chained together four separate vulnerabilities
• Successfully executed the attack
This isn’t just automation, it’s weaponized intelligence capable of finding weaknesses that human experts missed for decades.
Zero-Day Vulnerabilities Discovered
In controlled testing, the model uncovered thousands of previously unknown vulnerabilities, including long-standing flaws buried deep in widely used systems.
Among the most alarming findings:
• A 27-year-old vulnerability in OpenBSD enabling remote system crashes
• A 16-year-old flaw in FFmpeg capable of triggering memory corruption
• A 17-year-old FreeBSD exploit allowing full remote code execution
In one case, the AI independently constructed a complex exploit chain that older systems could only produce with heavy human assistance.
Explosive Leap in Hacking Capability
The scale of improvement over previous AI models is staggering.
In testing against the Firefox web browser:
• Older AI models produced 2 working exploits
• Mythos generated 181 working exploits
In broader system testing:
• The model created hundreds of crash scenarios
• Identified multiple full system takeover pathways
• Delivered results at extremely low cost, sometimes under $50 per exploit
Experts warn that this dramatically lowers the barrier for cyberattacks.
Big Tech and Banks Already Involved
Because of the risks, Mythos is not publicly available.
Instead, it’s being distributed through a tightly controlled initiative called Project Glasswing, involving major players such as:
- Amazon Web Services
- Apple
- Microsoft
- JPMorgan Chase
Other financial institutions, including Goldman Sachs, Citigroup, Bank of America, and Morgan Stanley, are reportedly gaining access to test their systems.
Anthropic has committed up to $100 million in credits to accelerate defensive testing.
Pentagon Flags National Security Risk
The situation is further complicated by the Pentagon, which recently designated Anthropic as a U.S. supply-chain risk over concerns about how its AI could be used.
The War Department has given the company six months to transition services, while Anthropic is challenging the designation in federal court.
Despite that warning, federal agencies are now actively trying to access the same technology—highlighting the urgency and uncertainty surrounding its capabilities.
A New Cyber Threat Landscape
Officials are increasingly concerned that AI like Mythos could:
• Discover vulnerabilities faster than they can be patched
• Automate large-scale cyberattacks
• Be weaponized by hostile actors
Anthropic itself has acknowledged the danger, warning that models like Mythos could enable “sophisticated cyberattacks” if not properly controlled.
The Next Generation of Cyber Warfare
The emergence of Mythos marks a turning point.
For years, cybersecurity has been a race between defenders and attackers.
Now, with AI capable of uncovering decades-old vulnerabilities in minutes, that balance may be shifting fast.
And as federal agencies, banks, and tech giants scramble to get ahead of the threat, one reality is becoming clear:
The next generation of cyber warfare may not be fought by humans at all, but by machines that can break into everything.
Be the first to comment