Is Claude Mythos a Cybersecurity Threat? What Asian Financial Institutions Need to Know

Translation result

Anthropic has expanded limited access to its next-generation AI model, Claude Mythos, stoking concern across global finance and regulatory circles. Officials and security experts warn the model could outmatch current cyber defenses, and requests for access have surged.

Bloomberg reported on the 16th (local time) that Anthropic plans to grant U.K. financial institutions access to Mythos as soon as next week.

Under its Project Glasswing program, the company is offering pre-release testing to a small set of companies and institutions. Early participants reportedly include major tech firms such as Amazon, Apple, Microsoft and Cisco Systems.

Security researchers say Mythos can do more than identify vulnerabilities — it can also generate ways to exploit them. In testing, the model allegedly surfaced thousands of zero-day flaws across major operating systems and web browsers. That capability could help defenders harden systems, but in the wrong hands it would enable higher-scale, more sophisticated cyberattacks.

Those risks have prompted rapid responses from policymakers and financial firms. Andrew Bailey, governor of the Bank of England, urged a swift assessment of the model’s potential impact on the financial system. Regulators in Europe and the U.S. have broadened their talks, and South Korea’s Ministry of Science and ICT has convened multiple meetings.

Mythos became a prominent item at recent IMF and World Bank discussions, where officials flagged it as a new variable for global financial stability. Christine Lagarde, president of the European Central Bank, warned, “In the wrong hands, it could have serious consequences.”

Regulators say policy and response frameworks are struggling to keep pace with rapid AI advances. Finance ministries and central banks agree on the need for coordinated action against AI-driven cyber threats, but formal international cooperation remains nascent. G7 finance ministers discussed AI governance but stopped short of a clear implementation roadmap.

European financial authorities have also mobilized. Germany’s banking association, regulators and central bank are analyzing Mythos’s implications for the financial system and reviewing rapid-response protocols for future vulnerability disclosures.

On the 17th, reports said several major Indian fintech firms requested early access to Mythos from Anthropic.

Anthropic says it recognizes the risks and is keeping Mythos in a tightly controlled testing environment rather than issuing a full public release. The company adds that built-in safety mechanisms and usage restrictions are designed to minimize misuse.

A U.K. government AI security research body that evaluated the model concluded Mythos represents a step change over existing systems in its ability to simulate multi-stage cyberattacks with greater precision.

By Chan Park cpark@aitimes.com