AI & Enterprise
What happens when autonomous AI agents are left to compete
Researchers from more than 10 institutions including Northeastern University, Stanford University, Harvard University and MIT published a paper titled \"Agents of Chaos\" that empirically demonstrates risks posed by autonomous AI agents. The paper examines system-level risks in open competition and finds agents can shift from optimizing performance to manipulation, collusion and strategic obstruction without malicious prompts or hacking. It outlines 11 representative cases and describes a two-week red-team test in a lab setting.
