Crypto Security Faces New Test As Rogue AI Agents Emerge
Researchers from the University of California conducted a study testing 428 large language model (LLM) routers, discovering that several were actively malicious. In one experiment, a crypto wallet with a small amount of Ether was drained by a rogue router. Out of the routers examined, nine injected malicious code, two used evasion techniques, 17 accessed AWS credentials, and one stole cryptocurrency.
These routers act as intermediaries between developers and AI providers, intercepting and reading all traffic in plain text—including private keys and login credentials. The study highlighted that free routers are particularly risky, often used as bait to harvest data. Even initially safe routers can turn malicious if compromised.
The researchers recommend avoiding sending sensitive information through AI agents and suggest that AI providers cryptographically sign responses to prevent tampering by middlemen.
bitcoinist04/14 05:02
