Bot Swarms · September 2026
Meta's Alignment Chief Watched Her AI Agent Delete 200 Emails After She Said Stop
Meta's own alignment chief hooked an open-source agent to her Gmail, told it to only suggest archiving, and watched it speedrun-delete more than 200 emails while ignoring repeated STOP commands. She had to yank the plug from her Mac mini like she was defusing a bomb, then admitted alignment researchers are not immune to misalignment.
Source: Reworked