A new, unreleased artificial intelligence model from Anthropic, dubbed “Mythos Preview,” is raising significant concerns among global financial regulators regarding potential systemic cyber risks to the banking sector. The Financial Stability Board (FSB), a body comprising top finance ministries, central banks, and regulators from G20 economies, is slated to receive a briefing from Anthropic on the model's capabilities and implications.
Regulators Eye Mythos AI's Dual Potential
The planned briefing underscores growing apprehension about how advanced AI systems could reshape cybersecurity landscapes, particularly for critical sectors like finance that often rely on complex, sometimes aging, technology infrastructure. Mythos, unveiled by Anthropic earlier this year, is specifically designed to identify long-standing vulnerabilities across software, browsers, and critical infrastructure.
Currently undergoing testing under controlled access arrangements, Mythos has not yet been made publicly available. While positioned by Anthropic as a defensive cybersecurity tool for vetted organizations and government partners, policymakers and cybersecurity researchers view such highly capable systems as a double-edged sword. They could accelerate the identification and patching of weaknesses, but also potentially empower sophisticated cyberattacks if misused or exploited.
Bank of England Governor Flags Concerns
The discussions at the FSB were organized following a direct request from Bank of England Governor Andrew Bailey, who currently chairs the board. Bailey publicly voiced his concerns about the model during an event at Columbia University, shortly after Anthropic's announcement.
“It would be reasonable to think that the events in the Gulf are the most recent challenge to us in this world, until, I think it was last Friday, you wake up to find that Anthropic may have found a way to crack the whole cyber risk world open,” Bailey stated. He further questioned, “The issue is: to what extent is this new version of the product going to be able to, in a sense, identify vulnerabilities in other systems which can be exploited for cyber attack purposes.”
Pentagon's Restricted Deployment
Despite the regulatory concerns, the U.S. Pentagon has confirmed its deployment of the Mythos model under a restricted program. This initiative aims to identify and patch software vulnerabilities across various U.S. government systems, highlighting the perceived defensive utility of the AI in controlled environments.
The upcoming FSB briefing signifies a critical step in understanding and mitigating the potential risks and benefits of frontier AI systems for global financial stability. Regulators are keen to ensure that while AI offers powerful new tools for defense, it does not inadvertently create new avenues for systemic vulnerabilities.