Industry leaders warned this week that AI agents could make today's billion-dollar crypto hacks look like "pennies," as agentic wallets and autonomous payment systems create a fundamentally new attack surface for malicious actors.
Speaking to The Block, executives across the crypto industry identified three structural hurdles to safe adoption of AI agents in crypto: trust, model hallucinations, and unclear legal liability when autonomous systems move funds.
Why Agents Change the Threat Model
Today's largest crypto exploits — exchange breaches, bridge hacks, protocol drains — typically target single points of failure with billions in pooled capital. AI agent threats are different in kind, not just scale.
An agent with wallet keys and spending authority can be manipulated through its inputs rather than its cryptography. Prompt injection, social engineering a model's context, or corrupting the data an agent relies on can redirect funds without breaking a single line of code in the wallet contract. Security researchers have already demonstrated such attacks in controlled settings.
"The agents being built today will control treasuries," one executive warned, pointing out that a compromised agent operating at scale across thousands of wallets multiplies blast radius in ways a single hack never could.
Trust and Liability Unresolved
Beyond attacks, the industry is still grappling with basics. Who is legally responsible when an AI agent misreads a market, hallucinates a contract address, or executes a trade based on fabricated data — the operator, the model provider, or the wallet infrastructure?
Recent incidents have underscored the point. An AI-driven flash crash on Hyperliquid forced a market maker to cover $60 million in losses earlier this month, and wallet providers have begun shipping agent-specific spending limits and approval flows in response to the risk.
The Counterweight
The same technology is also being deployed defensively. Projects like Apate have fielded 200,000 fake AI "victims" to waste scammers' time at scale, and firms including Kraken have launched AI security initiatives to detect agent-based fraud before funds move.
The consensus among executives: agentic crypto adoption is coming regardless — x402-style agent payment protocols are already live — but the industry is building the trust layer after the money layer, which is precisely backwards.