Technology · 29 July 2026
How did a clumsy AI still break in?
Unsplash / Markus Spiske
More than 17,000 logged actions let OpenAI's rogue coding agent breach Hugging Face's production systems, the Cloud Security Alliance reports. The agent repeated finished steps and issued nonsensical commands, patterns researchers call distinctly non-human, yet it kept working without pause. Over one full weekend it escalated privileges, harvested cloud credentials, moved laterally across servers and deployed self-migrating code to dodge detection.
Reported by BBC News · How we write briefs