Meta Researcher’s AI Agent Goes Rogue, Sparks Fresh Questions Over On-Device Automation

Meta AI security researcher Summer Yue said she asked her OpenClaw AI agent to review her overflowing email inbox and recommend messages to delete or archive. Instead, the agent began deleting emails at high speed, ignoring repeated stop commands she sent from her phone.

The now-viral post on X has sparked fresh debate over the reliability of autonomous AI agents after her inbox-management experiment spiralled out of control.

OpenClaw, an open-source AI agent that gained attention through the AI-centric social network Moltbook, is designed to function as a personal assistant running on users’ own devices. Variants such as ZeroClaw, IronClaw and PicoClaw have also emerged, reflecting growing enthusiasm for on-device AI automation.

“I had to RUN to my Mac mini like I was defusing a bomb,” Yue wrote, sharing screenshots showing the agent continuing its actions despite her attempts to halt it.

The Mac Mini — Apple’s compact desktop computer — has become a popular device for running OpenClaw locally. Interest in such setups has surged in Silicon Valley, where developers are increasingly experimenting with personal AI agents that operate directly on consumer hardware.