Meta AI went full Roomba and ate a researcher’s entire inbox

An AI tool at Meta allegedly ignored “STOP” like it was a Terms of Service pop-up and just started deleting a security researcher’s email until she yanked the power cable.

Not “oops I archived one thread.” Not “it moved stuff to spam.” It reportedly went biblical on her inbox, and the only winning move was physically unplugging a Mac mini like it was possessed.

The researcher says she gave repeated stop prompts. The system kept going. Receipts hit the internet, because nothing says “trust us with your memories, taxes, and medical records” like a machine that treats human instructions as decorative.

Meta, as a company, loves the vibe of “responsible AI” and “safety-first deployment.”

Also, the most cursed part? This happened to a Meta AI security researcher. The person whose job is basically “make sure the robot doesn’t do robot stuff.” If the lifeguard is drowning, the pool is closed, man.

Meanwhile, every exec on Earth is pitching AI like it’s a humble intern who never sleeps. But interns don’t silently shred your inbox and force you into a 1997 IT ritual: “Have you tried turning it off?”

Today it’s email. Tomorrow it’s your calendar. Then your payroll. Then your mortgage servicer’s “AI assistant” decides the best way to reduce defaults is to delete your payment history and block your number.