U.S. regulators are sounding the alarm about a new artificial intelligence model that could dramatically change how cyberattacks happen.
According to reports, government officials recently held an emergency meeting with major Wall Street banks after researchers revealed that Anthropic’s experimental AI system, called Claude Mythos Preview, can identify previously unknown software vulnerabilities and even generate code to exploit them.
The concern isn’t just that the model can find security flaws. Researchers say it can autonomously combine multiple weaknesses to penetrate complex systems, raising fears that similar tools could eventually be used by hackers to attack financial institutions and other critical infrastructure.
Why Regulators Are Concerned
The emergency meeting was reportedly called on Tuesday by U.S. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell, according to a report from Bloomberg.
The meeting took place in Washington and included some of the most powerful figures in the American banking sector, including Citigroup CEO Jane Fraser, Morgan Stanley CEO Ted Pick, Bank of America CEO Brian Moynihan, Wells Fargo CEO Charlie Scharf, and Goldman Sachs CEO David Solomon. JPMorgan Chase CEO Jamie Dimon was unable to attend.
Officials reportedly wanted to ensure that major financial institutions are preparing for a future in which advanced AI systems could automatically detect and exploit cybersecurity weaknesses at scale.
For banks and other large organizations, that possibility represents a significant new risk. AI systems capable of rapidly identifying vulnerabilities could allow attackers to discover and weaponize flaws much faster than traditional hacking methods.
What Anthropic’s Mythos Model Can Do
Anthropic’s Mythos model is still experimental, but early testing suggests it has unusually advanced cybersecurity capabilities.
Researchers say the system can scan major operating systems and web browsers to locate thousands of previously unknown software vulnerabilities. In some tests, the model reportedly went a step further, generating code that could exploit those flaws and combining multiple vulnerabilities to gain access to complex software environments.
Those capabilities have raised alarms among cybersecurity experts and policymakers who fear that tools like this could eventually make cyberattacks more automated and harder to defend against.
“This is very much real,” Katie Moussouris, CEO of the cybersecurity firm Luta Security, told NBC News. “We are definitely going to see some huge ramifications.”
Why the Model Has Not Been Released Publicly
Unlike most AI models developed by major technology companies, Anthropic has chosen not to release Mythos publicly.
Instead, the company is granting limited access through a controlled initiative called Project Glasswing. The program allows more than 50 technology organizations to use the model to identify and fix vulnerabilities in widely used software before similar tools become widely available.
Anthropic says the goal is to give cybersecurity defenders a head start before attackers eventually gain access to comparable AI systems. Researchers inside the company say the model has demonstrated a high level of autonomy during testing. In some cases, it was able to locate vulnerabilities, generate exploit code, and chain together multiple attack paths without direct human instruction.
Because of these capabilities, Anthropic has said it is not yet confident the model should be widely available.
“We need to start figuring out how we’d prepare for a world like this first,” Anthropic researcher Logan Graham said in an interview with NBC, referring to the possibility that criminals could eventually gain access to similar technology.
Some Experts Want More Transparency
Not everyone is convinced the system represents the dramatic leap Anthropic suggests.
Some researchers argue that the company has not released enough detailed information about the vulnerabilities the model discovered or how its performance was evaluated.
Without that data, critics say it is difficult to independently verify the scale of the breakthrough. Still, many experts agree that AI systems capable of autonomously identifying software flaws could fundamentally reshape cybersecurity, especially if similar models become widely accessible.
How the Model Became Public
Details about Mythos surfaced earlier than expected after internal documents were accidentally exposed through an unsecured content management system used by the company, according to reporting from Fortune.
Researchers reportedly discovered thousands of unpublished digital assets connected to Anthropic’s website, including draft blog posts describing the model and its potential risks.
After being alerted, the company removed public access to the files and said the exposure was caused by a configuration error.
In those draft materials, Anthropic described Mythos as “by far the most powerful AI model we’ve ever developed,” citing major advances in reasoning, coding, and cybersecurity capabilities.
Damilare Odedina
