AI agents could make today's billion-dollar crypto hacks look like 'pennies,' industry leaders warn

Quick Take

• Panelists see trust, hallucinations and unclear legal liability as the biggest hurdles to agentic adoption.
• AI agents could lower friction across decentralized systems while introducing new privacy and security risks.

Artificial intelligence agents are being hyped up as the answer to eliminating the friction involved in navigating crypto and next-gen financial services, but industry heads warn they could do the same for bad actors looking for their next targets.

"We're acting like agents are always good, and I think that's a really fatal flaw in almost everything agentic today," Global Settlement Network CEO Ryan Kirkley said Wednesday during a panel moderated by The Block CEO Steve Chung at the Wyoming Blockchain Symposium 2026.

Kirkley said AI agents could allow hackers to break into Wi-Fi networks, passwords and wallets at a scale that has previously been impractical.

"We think these bridge hacks are bad," Kirkley said. "It's pennies. It's nothing."

"It used to be so hard to try to hack a person worth $20,000 because why would you spend all that time going after one person?" he added. "Now I can have an agent that goes after everyone."

Web3 Foundation Vice President of Technical Operations Bill Laboon agreed, saying that the efficiencies that make decentralized systems easier to navigate cut both ways.

"This also means that there is less friction for the bad guys," Laboon said. "For the good guys, the bad guys and the neutral guys."

Who controls the agent?

The panelists also debated how much authority users should hand over to their AI agents.

Midnight Foundation President Fahmi Syed argued that agents need clearly set parameters, not unrestricted access to all of a user's personal and financial information.

"The concept of giving all that power to a single agent to go and access my credit card information, my personal security, Social Security information, all my personal information, as well as access to all my different accounts, is a scary concept without putting in parameters," Syed said.

Kirkley sees setting an agent's authority as one of the easier problems to solve, adding that securing the underlying system is a much bigger concern.

"Are we not creating an attack vector where an agent could be taken over and drain an entire wallet?" Kirkley said.

Laboon also raised concerns around privacy, saying that even with systems designed to keep transactions private, metadata leaks could potentially be used by AI to piece together information that users thought was secure.

"This is really what I'm concerned about: these metadata leaks that people are going to think that they're private because they're using a private chain," Laboon said.

Trust and adoption

Then there is the issue of trust, which the panelists mostly agreed was one of the biggest obstacles to mainstream adoption of AI agents.

Laboon brought up the tendency of large language models to hallucinate as a reason why most would feel uneasy giving an agent control of their finances.

"My LLMs still occasionally hallucinate," Laboon said. "I wouldn't want to put my 401k in the hands of that."

Silvermine Capital Advisors founder Richard Shorten similarly said that agentic AI tech has "moved further and faster" than people have been able to process.

"The question is getting through the trust," Shorten said. "It's sort of turning that toolkit, that capability, into something where I trust to let it loose on portions of my business or my life."

Kirkley said regulations could prove to be another major hurdle, especially ironing out who takes responsibility when autonomous agents make mistakes or even break the law.

"If your agent goes and does something illegal, or does something where it spends money that you can't have, how do you claw it back? Where is the ultimate decision maker?" Kirkley said. "It's all about where does the actual buck stop, and where do we hold the legal liabilities?"