95% of global markets face AI cybersecurity risks — can regulators keep up?

Global financial regulators are racing to get ahead of a problem they helped create. As artificial intelligence reshapes everything from fraud detection to trading algorithms, it is also generating a new category of AI cybersecurity risks that watchdogs are only beginning to fully reckon with. Switzerland’s top market regulator is now leading a coordinated international push to ensure that regulators, not just the banks they oversee, are armed with the right tools.

Key takeaways

• FINMA president Marlene Amstad is calling for rapid technology adoption in financial institutions to counter accelerating AI-driven cyber threats.
• FINMA co-created an international AI supervisory forum under IOSCO, covering authorities responsible for roughly 95% of global financial markets.
• Around 100 policy and technology experts gathered at a recent hackathon to build new crypto-market supervision tools.
• The US government ordered Anthropic to suspend exports of its Mythos and Fable AI models this month, citing national security concerns.
• Chinese firm 360 Security Technology responded by announcing a domestic alternative to the Mythos model.

FINMA Leads International Forum to Promote AI in Financial Regulation

Switzerland’s Financial Market Supervisory Authority has done more than flag the problem — it helped build the infrastructure to address it. FINMA played a central role in establishing a dedicated forum within the International Organization of Securities Commissions (IOSCO), the standard-setting body for global market regulation, specifically to accelerate AI adoption among financial watchdogs.

The reach of that forum is substantial. It connects supervisory authorities that collectively oversee approximately 95% of global financial markets, making it one of the most expansive regulatory coordination efforts in recent memory. The ambition is clear: rather than waiting for AI threats to escalate, regulators want to develop shared tools and shared understanding before they are caught off guard.

Marlene Amstad — president of FINMA and chair of the forum — has positioned the initiative as a matter of urgency, not long-term planning. Speaking to Reuters, she framed the challenge in blunt operational terms: “As hackers move faster, banks must adapt by patching vulnerabilities more rapidly.” That sentiment drives the logic of the entire forum.

Accelerating Adoption of AI Tools to Address Rising Cybersecurity Risks

The clearest demonstration of that urgency came this week. Around 100 policy and technology specialists convened for a hackathon — an intensive, collaborative work session — with one specific goal: building new tools for crypto-market supervision.

The event brought together experts from across the policy and technology divide, a combination that is still relatively rare in regulatory circles. Crypto markets, which operate continuously and across jurisdictions, present some of the most complex supervisory challenges in modern finance. The idea of regulators building their own monitoring tools — rather than procuring them from outside — signals a meaningful shift in how watchdogs are thinking about their own capabilities.

Amstad also flagged that regulators are examining whether safeguards can be embedded directly into digital asset systems, rather than applied on top of them after the fact. That architectural approach would represent a more structural response to financial sector security — hardening the infrastructure itself rather than patching problems as they surface.

Emerging AI-Related Cybersecurity and National Security Challenges

When AI models become the vulnerability

The urgency behind these efforts is partly traceable to a concrete and uncomfortable development: advanced AI models have themselves become a source of exposure. Experience with models including Anthropic’s Mythos has surfaced software vulnerabilities that carry real operational and security implications for financial institutions, according to Amstad.

The US government moved quickly on that concern. This month, Washington ordered Anthropic to suspend exports of its latest Mythos and Fable AI models, invoking national security as the basis for the restriction. It is a striking intervention — one of the most high-profile moves yet to treat cutting-edge AI model access as a strategic asset requiring export controls, in the same way advanced semiconductors have been treated in recent years.

China’s rapid domestic response

The geopolitical reaction was equally swift. Chinese cybersecurity firm 360 Security Technology announced this week that it has developed a domestic alternative to the Mythos model. The timeline — a domestic substitute surfacing almost simultaneously with the US export ban — illustrates just how compressed the competitive cycle around frontier AI has become.

For regulators like FINMA, this creates a difficult balancing act. Export restrictions on the most advanced models could limit the very tools that watchdogs are trying to use to strengthen financial system resilience. Amstad addressed this directly, stating that “Switzerland must retain access to the most advanced AI models,” and that AI will be instrumental in toughening up financial systems before they are deployed in the field.

Regulatory Strategies for Enhancing Financial Sector Resilience

The broader strategic picture emerging from FINMA’s approach reflects a regulator that has internalized a core tension: AI is simultaneously the threat vector and the solution. The same capabilities that make AI models useful for fraud detection, anomaly monitoring, and systemic risk assessment are the capabilities that, when turned toward malicious ends or left with unpatched vulnerabilities, represent genuine dangers to financial stability.

Rapid vulnerability patching is one front. Embedding safeguards into digital asset architecture is another. Building shared supervisory tools through collaborative events like the recent hackathon is a third. And maintaining access to frontier AI models — rather than accepting restricted access — is the fourth pillar Amstad has publicly championed.

What makes the FINMA-led IOSCO forum particularly significant is its scale. Coordinating across authorities that represent 95% of global financial markets means that any tools or frameworks developed through this network carry real systemic weight. A supervisory gap in one jurisdiction is far less likely to become a contagion point if the broader regulatory community is operating from a shared, AI-assisted baseline.

The question that follows — and one the industry will be watching closely — is whether that coordination can move fast enough. If hackers and adversarial actors are already operating at machine speed, the test for this generation of regulators is whether international collaboration can compress its usual timelines to match.

FAQ

Why is FINMA emphasizing rapid adoption of new technology in financial institutions?

FINMA president Marlene Amstad highlighted that hackers are moving faster, requiring banks to patch vulnerabilities more rapidly to mitigate AI-driven cyber risks. Speed of response is now central to financial sector security strategy.

What is the scope of the international forum created by FINMA under IOSCO?

The forum promotes AI adoption among supervisory authorities covering about 95% of global financial markets, making it one of the broadest regulatory coordination mechanisms currently active in financial oversight.

What recent collaborative event did regulators hold to develop supervisory tools?

Around 100 policy and technology experts participated in a hackathon aimed at building new tools for crypto-market supervision, combining regulatory expertise with technical development in a single collaborative setting.

How have AI models like Anthropic’s Mythos impacted cybersecurity concerns?

Experience with such AI models has exposed software vulnerabilities that increase cybersecurity and national security risks — a factor that led the US government to order Anthropic to suspend exports of its Mythos and Fable models this month, citing national security concerns.

Article produced with the assistance of artificial intelligence and reviewed by the editorial team.