Researchers are warning the crypto industry to treat AI agents as untrusted systems — a shift in mindset that could head off a wave of financial losses as automated bots handle more transactions. The call comes as developers race to integrate large language models and autonomous agents into trading, wallet management, and DeFi protocols.
The core warning
The message is straightforward: assume an AI agent can be compromised at any moment. Treating these systems as untrusted forces developers to build with security from the ground up rather than bolting it on later. Researchers argue that without this default posture, the financial vulnerabilities created by AI agents — which can execute transactions faster than any human — will multiply.
Crypto transactions are irreversible. A compromised agent that sends funds to the wrong address or approves a malicious contract leaves victims with little recourse. The risk isn't theoretical. Automated trading bots and AI-powered wallet assistants are already live on major networks, handling sums that would make a human trader flinch. If an attacker takes control of one of these agents, the damage can happen in seconds.
What kind of measures
Researchers recommend a bundle of practices that treat the agent as a potential adversary. That means sandboxing its execution environment, limiting its access to only the smart contracts and balances it absolutely needs, and requiring multi-signature approval for high-value actions. Real-time monitoring and kill switches should be standard. The idea is to contain any breach before it spirals.
The warning arrives as AI agents move from experimental to operational in crypto. The same technology that lets them parse natural language and make decisions also makes them opaque — hard to audit and easy to exploit. Treating them as untrusted isn't pessimism; it's basic engineering hygiene. Developers building AI-powered crypto tools are being urged to start there — or risk finding out the hard way why the researchers are right.
