TRM Labs: For AI Agents Moving Money, Vet the Human Behind Them

Verify the human behind the AI agent. That's the short version of TRM Labs' answer to the question every bank is about to face: who's on the hook when software moves the money? Speaking at Merge Madrid, TRM Labs executive Guillaume Rodenas said checking the principal — the person holding the mandate that controls the agent — is the place to start. From there, the agent's transaction history fills in what it did on that person's behalf.

The big picture: Binance lost roughly $6.3 billion last week to North Korean hackers who talked their way in through a fake job interview, then exploited an internal AI tool, according to BeInCrypto. Most of the haul was gone inside 20 minutes.

The numbers behind the scare are brutal. TRM Labs estimates North Korea accounts for 85% of the value stolen in crypto hacks this year, with total losses hitting a record $8.7 billion — largely driven by Binance. Rodenas said criminals are now running autonomous agents that poke at protocols and fire off transactions around the clock, and warns AI is creeping into attacks on conventional banks too. His conclusion: if the crime runs 24/7, the defenders have to as well.

Accountability gets murkier from there. Rodenas floated that developers could carry responsibility for software bugs, while users still need to understand the financial protocols they're touching. DAOs make the question harder still — he tied that uncertainty to why Europe's MiCA rules have largely left fully decentralized finance outside their reach.

TRM already leans on AI to monitor wallets and flag suspicious moves. The bottleneck, Rodenas said, sits with European banks, where readiness varies but willingness varies more. Some have dedicated blockchain teams; others have staff still trying to convince management that digital assets are worth a second look.

Rodenas thinks the crypto label itself is on borrowed time. In a few years, he predicted, the industry will drop the word entirely and just call it money. The catch for banks: that future means building for a system where software can move funds without pausing to ask a human first.