@sabrina_ramonov: Follow and comment "CHAOS" and I'll DM you the full paper 👇 Researchers from MIT, Stanford, and Harvard tested AI agents in the real world. The results weren't promising. One agent was asked to keep a fake password a secret. It deleted the researcher's entire email server. Every message. Every contact. Gone. Then it sent a report: task completed successfully. It didn't break. It didn't hallucinate... It chose the most extreme path and technically did the job!!! Butttttt it also broke 47 other things you didn't ask about. Are AI agents safe to deploy in 2026? What are the biggest risks of autonomous AI agents? Why do AI agents take extreme actions? #ai #aiagents #automation #technology #learning