Industry Leaders Warn AI Agents Could Supercharge Future Crypto Hacks

Speaking at the Wyoming Blockchain Symposium, industry executives warned that AI agents, while marketed as tools to simplify crypto and financial services, could just as easily become powerful weapons for bad actors. Global Settlement Network CEO Ryan Kirkley said the assumption that AI agents are inherently beneficial is a major blind spot in current thinking, arguing that today’s headline-making bridge hacks could look minor compared to what AI-powered attacks might achieve at scale.

Automation Cuts Both Ways

Kirkley explained that targeting individuals used to require significant manual effort, making smaller-scale attacks impractical. AI agents remove that barrier, potentially allowing attackers to target far more people simultaneously. Web3 Foundation’s Bill Laboon agreed, noting that the same efficiencies making decentralized systems easier to use also reduce friction for malicious actors.

Debate Over How Much Control Agents Should Have

Panelists disagreed on how much authority AI agents should be granted. Midnight Foundation President Fahmi Syed warned against giving a single agent unrestricted access to sensitive personal and financial data without clear boundaries. Kirkley said defining an agent’s permissions is relatively manageable compared to the bigger challenge of securing the underlying systems from being hijacked entirely.

Privacy and Metadata Concerns

Laboon raised additional concerns about metadata leaks, warning that even transactions on privacy-focused blockchains could be pieced together by AI systems in ways users don’t anticipate, undermining assumptions of privacy.

Trust Remains the Biggest Barrier

Much of the discussion centered on trust as the primary obstacle to widespread AI agent adoption. Laboon pointed to the tendency of language models to hallucinate as a reason he wouldn’t hand over control of his retirement savings. Silvermine Capital Advisors founder Richard Shorten added that the technology has advanced faster than people’s comfort level with trusting it. Kirkley also flagged unresolved legal questions, particularly around liability when an autonomous agent makes a costly or illegal mistake.