When a trading agent can sign on-chain transactions without needing human approval for every single one, the fallout from a single model hallucination could be far more than just a transfer sent to the wrong wallet.
According to The Block, crypto industry figures are warning that if AI agents are granted greater autonomy to execute on-chain operations, the losses resulting from flawed decisions or malicious exploitation could dwarf even the billion-dollar hacking incidents seen in the past. As the report puts it, today's headline-making cases could end up looking like "pennies" by comparison.
The report identifies three major obstacles currently blocking the widespread adoption of AI agents in crypto: the lack of established trust between users and institutions toward these agents, the persistent tendency of large language models to "hallucinate" and produce flawed judgments, and the unresolved question of who bears legal liability when an autonomously operating agent makes a costly mistake.
All three concerns point to the same core issue: as decision-making shifts from humans to automated agents, the speed and scale at which risk can materialize also changes—and neither current trust mechanisms nor legal frameworks have caught up.
The report did not specify who made these remarks, when they were made, or details of any related incidents.
