New York, regtech standard, published September 10.

Elliptic, the on-chain analytics firm that has spent over a decade helping exchanges and banks track illicit crypto activity, published a new agentic crypto compliance standard it’s calling the Elliptic Standard: eight principles for governing financial crime risk when the transactions are being originated and settled by AI agents instead of people. The company says agentic crypto payment volume is up 522% since May, based on its own intelligence data, and it’s positioning this as the moment blockchain compliance tooling either adapts or falls behind the thing it’s supposed to be monitoring.

The core argument is that most on-chain compliance today is investigative: something suspicious happens, an analyst reviews it, a report gets filed, sometimes days or weeks later. That model assumes a human pace of transacting.

Elliptic’s case is that once agents are both sides of a transaction, initiating and countersigning at machine speed, a backward-looking review process can’t catch anything before the money has already moved. The eight principles it’s proposing cover things like model transparency (no black-box risk scoring), foundation-model agnosticism (don’t get locked into one AI vendor), and human override built in from the start rather than bolted on after a regulator asks about it.

Circle and CoinFlip both put compliance executives on the record supporting the framework, which is a meaningful signal since Circle in particular has skin in the game as a stablecoin issuer that needs its own compliance story to hold up under scrutiny. Whether other major players (Coinbase, Binance, the custody banks moving into digital assets) adopt anything resembling Elliptic’s specific eight points is a different matter. Standards proposed unilaterally by a single vendor, however well-reasoned, tend to succeed only when a critical mass of the industry treats them as the baseline rather than one company’s marketing document.

The stats underpinning the urgency are worth a second look before taking them at face value. The 522% agentic transaction growth figure is Elliptic’s own internal measurement over a three-month window, not an independently audited number, and “agentic crypto payments” is still a loosely defined category industry-wide. The $893.3 million FBI figure tied to an AI nexus in 2025 is real and sourced, but it covers all AI-linked fraud losses, not specifically agent-to-agent crypto transactions, so it’s doing more scene-setting than direct evidentiary work for Elliptic’s specific thesis.

Self-interest doesn’t make the underlying problem fake. Machine-speed transacting genuinely does break a compliance model built around human review timelines, and somebody in this industry was going to try to define the standard first. Elliptic has the on-chain data history and the existing customer relationships in the crypto compliance space to make a credible run at being that somebody, which is more than most vendors publishing a “standard” can claim.

What the Agentic Crypto Compliance Standard Actually Covers

The eight principles behind Elliptic’s agentic crypto standard focus on keeping a human in the loop even as transaction speed increases: transparent, explainable risk models instead of black-box scoring, the ability to swap out the underlying AI model without rebuilding the compliance stack, and override authority that’s built into the system architecture rather than added after a regulator flags a gap.

Elliptic has published its full breakdown of the standard on its company site. For more on how regulators and vendors are responding to AI-driven risk, see Fintechbits’ regulatory updates section. Fintechbits also covered a related AI product launch today in wealthtech, where FinTurk’s new PortfolioSolver tool raises similar questions about how much autonomy to hand an AI system before a human signs off.

The real test comes the next time a major exchange gets hit with an agentic-fraud incident large enough to draw regulatory attention. Whichever framework gets cited in that aftermath is the one that actually becomes the standard.