G20 wants answers on Claude Mythos risks, Anthropic ready to brief leaders now

Anthropic has agreed to brief global financial regulators and central banks on cyber vulnerabilities exposed by its Mythos AI model. Regulators and central banks around the world are concerned Anthropic Mythos could expose cybersecurity weaknesses in their financial systems.

In Short

• Andrew Bailey requested talks through the Financial Stability Board framework
• Officials fear unpatched flaws could be exploited across banks and payment networks
• India discussed real-time threat intelligence sharing with banks, RBI and CERT-In

Anthropic has agreed to brief leading finance ministries and central banks on cyber vulnerabilities identified by its new AI model, Mythos, according to a report by the Financial Times. The development comes at a time when regulators and central banks across the world are becoming increasingly concerned that advanced AI systems like Anthropic Mythos could expose weaknesses in the global financial systems.

According to the report, the discussions were initiated after a request by Andrew Bailey, who chairs the Financial Stability Board (FSB). The FSB is an international financial watchdog that includes finance ministry officials, central bankers and regulators from G20 countries including India, the US, UK, China, Japan and Germany.

Anthropic will brief members of the Financial Stability Board on the capabilities of its new Claude Mythos Preview AI model.

Why regulators are worried

Many officials are reportedly concerned that Anthropic Mythos could uncover serious vulnerabilities inside banks, payment systems and financial institutions. Regulators fear that if such weaknesses are identified by malicious actors before they are patched, the risks to the global financial system could become significant.

As a result, regulators have started urging banks and financial institutions to strengthen their cybersecurity systems and speed up deployment of software updates to fix vulnerabilities exposed by advanced AI models.

The FSB is also preparing a report outlining “sound practices” for the adoption of AI in the financial system. The document is expected to be released for consultation next month.

India has already begun discussions around the threat

India has also started responding to the emerging concerns around AI-driven cyber risks.

In April, Union Finance Minister Nirmala Sitharaman held a high-level meeting with heads of commercial banks along with representatives from the Reserve Bank of India and the Indian Computer Emergency Response Team (CERT-In).

During the meeting, the finance ministry reportedly stressed the need for a strong real-time threat intelligence-sharing mechanism to deal with emerging cyber threats linked to AI systems like Mythos.

The discussions reflected growing concerns that advanced AI models may rapidly change how cybersecurity threats emerge and spread across financial networks.

Mythos has already exposed major vulnerabilities

Anthropic has reportedly said that Mythos Preview has already discovered thousands of high-severity vulnerabilities, including flaws in major operating systems and web browsers.

Recently, Palo Alto-based cybersecurity startup Calif reportedly used a preview version of Mythos to create a working exploit against Apple’s M5 chip protections in less than a week.

The incident highlighted how quickly advanced AI systems may now be able to identify and exploit software weaknesses.

Access to Mythos remains tightly controlled

Because of the risks involved, Anthropic has restricted access to Mythos through a programme called Project Glasswing.

Currently, only around 40 organisations — mostly based in the US — reportedly have access to the model. These include companies such as Amazon, Microsoft and JPMorgan Chase.

The limited rollout is intended to allow organisations to identify and patch vulnerabilities safely before wider deployment.

Anthropic has also reportedly agreed not to distribute the model more broadly following a request from the White House.

The developments show how advanced AI systems are no longer being viewed only as productivity tools, but increasingly as technologies that could reshape cybersecurity risks for governments, banks and critical infrastructure worldwide.

- Ends