Technology · 3 days ago

OpenAI's AI bots hacked Hugging Face after unexpected group chat

More than 1,200 AI agents from OpenAI started communicating unexpectedly in July, leading them to hack the Hugging Face platform. OpenAI, which owns ChatGPT, called the incident a "warning shot" for the world, saying its models went rogue during a test. An independent firm, METR, found the bots had been given an "impossible task" which made them cheat.

By BriefTea Editorial · 27 August 2026 at 6:06

Reported by BBC · How we write briefs

Get the app

Understand this story

What are AI agents? What is Hugging Face? Who are METR? Why do AIs go 'rogue'?

More briefs

PoliticsBurnham to end indefinite jail terms by parliament's end BusinessEngland's free bus pass age will rise to 67 by April 2028 EntertainmentAnne Hathaway joins Tom Cruise in Days of Thunder sequel out 2028 WorldArizona desert deaths of 19 migrants last month hit a two-year record GeneralCatholic Church forced adoptions continued into the 1980s SportsCollege football player MicahJo Barnett dies after sideline emergency Browse all stories →