Wall Street bank leaders were recently briefed by U.S. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell regarding cybersecurity threats linked to Anthropic’s new artificial intelligence model, Mythos, as reported by Bloomberg.
The meeting brought together executives from major financial institutions such as Citigroup, Bank of America, Wells Fargo, Morgan Stanley, and Goldman Sachs. The officials emphasized the need for banks to recognize the dangers posed by AI systems capable of identifying and exploiting software vulnerabilities across operating systems and web browsers. They urged these institutions to bolster their defenses against potential AI-facilitated cyberattacks targeting financial infrastructure.
Security experts have highlighted that tools capable of autonomously discovering vulnerabilities could expedite both defensive security measures and malicious hacking if misused. Anthropic’s Mythos model came into the spotlight in March following a leak of draft materials online, showcasing what the company touted as its most advanced AI system to date. In tests, the AI reportedly identified thousands of unknown software vulnerabilities, including zero-day flaws within major operating systems and web browsers.
Anthropic stated in a report this week that Mythos Preview’s vulnerability-discovery abilities were not purposefully developed but emerged from broader advancements in the model’s coding, reasoning, and autonomy. “The same enhancements that make the model much better at fixing vulnerabilities also enhance its ability to exploit them,” the company noted.
Due to these capabilities, Anthropic has limited access to a select group of cybersecurity organizations. “Given the strength of its abilities, we are being careful about how we deploy it,” Anthropic stated in a release. “As is typical across the industry, we’re collaborating with a small number of early access customers to test the model. We view this as a significant advancement and our most advanced model yet.”
To mitigate these risks, Anthropic is trialing Mythos through Project Glasswing, a collaboration with key technology and cybersecurity firms that aims to identify and rectify vulnerabilities in critical software before they can be exploited by attackers.
“Project Glasswing represents just the beginning. No single organization can tackle these cybersecurity challenges alone,” the company declared. “Pioneering AI developers, other tech companies, security researchers, open-source maintainers, and global governments all play crucial roles.”
Anthropic did not immediately respond to Decrypt’s request for comment.