AI agents must be treated as untrusted systems: Researchers

A research paper co-authored by Google, Gray Swan AI, and others posits that current AI security focuses too heavily on model robustness. Instead, experts argue that AI agents should be treated as untrusted components within a broader computer security framework. To mitigate risks in the rapidly growing Web3 and crypto sectors—where autonomous agents are increasingly interacting with wallets and protocols—the study recommends three key mechanisms: separating instructions from untrusted data, enforcing the principle of least privilege, and strictly controlling data flow. These measures aim to prevent common vulnerabilities like prompt injection and unauthorized access to sensitive information.

(Source：Cointelegraph)