Researchers: AI Agents Should Be Treated as Untrusted Systems

Google and Meta researchers argue that AI agents interaction with external systems requires treating them as inherently untrusted to prevent misuse and ensure security.