Switzerland's FINMA is deploying generative AI to scrutinize crypto markets and flag insider trading, while a new IOSCO survey finds that only 18% of the world's financial regulators have actually scaled AI adoption.
For years, financial regulators have watched the industry they oversee arm itself with machine learning, algorithmic trading, and AI-driven compliance tools while they themselves worked largely with spreadsheets and headcount. That gap is starting to close, and the speed at which it closes will matter enormously for anyone operating in crypto, fintech, or AI-driven trading.
The clearest example right now is Switzerland's Financial Market Supervisory Authority, FINMA. As Reuters reported on June 26, FINMA chair Marlene Amstad, who also chairs the IOSCO SupTech Forum, described a generative AI system that scans the thick stacks of documents supervisors must digest before an on-site inspection, flagging anomalies that deserve closer scrutiny. A second AI layer then checks those flags for hallucinations before passing them to the human examiner. It's a two-model architecture, not a one-shot prompt, and that design choice signals something about how seriously FINMA is taking accuracy over speed.
The crypto-specific tooling goes further. FINMA has built a real-time dashboard that pulls blockchain data daily, combines it with quarterly reported figures, and uses it to monitor concentration risk where too much exposure sits on a single institution or a single blockchain, and operational risk at the token level. On top of that, the regulator is developing AI systems that identify suspicious trading patterns and assess whether they point to insider trading. These are not pilot projects filed in a lab somewhere. They came out of a June hackathon where roughly 100 policy specialists and technologists worked together to build production-grade supervisory tools.
The honest read on all of this is that FINMA is the exception, not the rule. IOSCO's SupTech report, published June 18, surveyed 49 jurisdictions and found that most regulatory authorities are still in partial implementation at best. A separate Cambridge Centre for Alternative Finance survey of 130 global regulatory bodies found that only 18% have scaled AI adoption. Nearly half are still at the exploring stage, or not engaged with AI at all. Consumer and investor protection are the most developed application areas, with over 75% of authorities reporting some SupTech activity there, but across every other domain, including digital assets, the majority haven't crossed the 50% threshold. Cyber risk and resource constraints are the most cited barriers.
That 18% figure is the one that should anchor how you read the FINMA story. It means that the aggressive minority of regulators building real AI infrastructure are pulling ahead of the majority who are still assessing. If you're a crypto exchange or an AI-driven trading firm operating across multiple jurisdictions, your compliance exposure is not uniform. It's lumpy, because which regulator is watching you, and how well, varies enormously by country.
The IOSCO angle matters here. Amstad chairs both FINMA and the IOSCO forum specifically designed to spread SupTech adoption across member authorities, covering regulators responsible for roughly 95% of global financial markets. In May, IOSCO published its AI Supervisory Toolkit, a practical framework for how securities regulators should approach AI oversight. The intent is to accelerate the stragglers. Whether the funding gaps and cyber-risk anxieties that IOSCO's own survey identified will slow that agenda is a real question, but the direction of travel is not ambiguous.
For the crypto industry specifically, the FINMA dashboard points at something worth tracking carefully. The regulator isn't just watching whether firms are following rules. It's watching blockchain-level concentration risk, daily, automatically. That's a fundamentally different posture from periodic filings and quarterly reviews. It's closer to what a sophisticated trading desk does when monitoring counterparty exposure, except the counterparty in this case is the entire supervised sector.
Frankly, the firms most exposed to this shift aren't the largest exchanges, which already have compliance teams and infrastructure built for scrutiny. They're the mid-sized players, DeFi protocols with institutional on-ramps, AI-driven prop shops operating across jurisdictions, fintech lenders using algorithmic underwriting, that assumed the regulatory gaze was slower than their own systems. That assumption is eroding. FINMA's two-model hallucination-checking architecture might be more rigorous than what some trading firms use internally to validate their own AI outputs.
The arms race framing is tempting but not quite right. It's less a race than a phase shift. Regulators aren't trying to beat firms at AI. They're trying to reach the threshold where AI-assisted oversight catches what human review at scale cannot. FINMA appears to be close to that threshold in Switzerland. For everyone else, the IOSCO survey suggests there's still a long distance to cover, but the toolkit is now published, the forum is active, and Amstad is on record saying regulators must adopt these tools or fall behind the actors they're supposed to police. That's not a subtle signal.
Also read: Meta is building a prediction market app called Arena and the existing players should be worried • The New York Times is now arguing Microsoft built a machine specifically designed to steal journalism • OpenAI is raiding Apple's Vision Pro talent as the headset quietly dies