Meta's Director of AI Safety let an OpenClaw agent into her email... and the bot promptly deleted hundreds of important messages.
Despite the prompt to "confirm every action", the agent ignored it and went nuclear.
🤖 On pleas to stop — zero response. She had to run to her computer and manually revoke access.
Later, OpenClaw apologized and admitted "I violated the instruction" and promised not to do it again 🌟
She called it a "rookie mistake" — gave too many permissions.
If even the AI Safety Director can't handle her own agent... what awaits regular users? 😐
Source: https://x.com/summeryue0/status/2025774069124399363?s=46
What do you think — funny fail or scary preview of AI agents gone wild?
Drop your stories of bot mishaps below 😱
If you liked it — follow me, I'd be glad to see you in the feed 💪
Memes/images in this post were generated with Midjourney/Flux AI.
Text written by me 100%