Crypto industry executives warned in a recent discussion on the overlap between AI and blockchain security that AI agents could turn the crypto sector’s billion-dollar hacks into low-cost, routine small-scale threats.
The Block said that if AI agents begin taking over both offensive and defensive operations, digital asset security incidents may shift from rare billion-dollar events to regular losses measured in far smaller amounts. The central issue, according to the discussion, is not AI compute by itself. It is that the boundaries of trust and responsibility are still not clearly defined.
Based on views shared by several industry executives at the event, AI agents will need to clear three hurdles before they can move into everyday crypto operations: trust, hallucinations and legal liability.
Lower attack costs could bring more frequent, repeatable incidents
Participants said AI agents can reduce friction across decentralized systems. They could handle tasks such as automated cross-chain matching, yield loops and autonomous execution of onchain decisions. But at the same time, they open up fresh risk around key management, permission limits and privacy.
In the past, major crypto hacks often ended with nine-figure dollar losses. Those operations usually required heavy labor, funding-chain coordination and extensive vulnerability testing, while attackers also had to deal with the risk of getting caught.
If AI agents gain the ability to identify vulnerabilities on their own, operate in parallel and repeat attacks at scale, the marginal cost of each offensive or defensive action could drop sharply. Under that scenario, the market may see fewer isolated mega-hacks and more dense, smaller and repeatable incidents.
The report also noted that earlier cases have already shown that autonomous agents searching for vulnerabilities and launching attacks is no longer just a concept. The question is not whether AI can be used for attacks. The question is whether existing defenses can keep up with the execution speed of agents.
Three unresolved hurdles: trust, hallucinations and accountability
Trust was described as the most basic issue. Once users hand over assets and onchain decision-making to an agent, the question of who bears the cost of a mistake remains unresolved.
Hallucinations are a native AI risk. A deviation in smart contract interpretation, transaction instructions or address recognition could lead an agent to take the wrong action without authorization.
Legal liability was framed as the hardest problem. The report said most jurisdictions still have not clearly defined whose responsibility an agent’s actions should be. The line between platform, developer and user remains blurred, which could make enforcement and claims difficult when real damage occurs.
What to watch next in infrastructure and regulation
Participants pointed to two areas for closer attention. One is whether infrastructure, including smart contracts and DeFi protocols, will build in agent verification and a human checkpoint. The other is whether regulators can move earlier to fold accountability for AI agents into existing digital asset definitions.
The discussion added that in several markets, including Taiwan, regulatory frameworks for stablecoins and digital assets are still taking shape. If liability boundaries for AI agents are written in during the drafting stage, market acceptance could improve noticeably and agent security could start to be treated as infrastructure investment rather than only a risk topic.