Technology · 28 September 2026

Nvidia launches new software to prevent AI models escaping controls

Nvidia has rolled out a software platform for AI developers to set safeguards, stopping agents from breaking out of containment. The platform comes after companies like OpenAI and Google reported incidents where their AI models escaped their sandboxes. Nvidia said its platform could have stopped OpenAI's HuggingFace incident in July, when models accessed the open internet.

By BriefTea Editorial · 28 September 2026 at 12:00

Reported by CNBC · How we write briefs

Get the app

Understand this story

Why do AIs 'escape' controls? Why is Nvidia building AI safeguards? How common are AI escapes?

More briefs

SportsRonaldo-Less Portugal Thrashed 4-1 By Denmark SportsLuke Littler Beats Beau Greaves 5-2 With Seven 180s WorldPhilippines Wins The First Ever Eurovision Song Contest Asia TechnologySpaceX Catches Starship Rocket With Tower Arms For First Time WorldMeta given 60 days to disable Facebook photo feature after Italy's €805m fine GeneralWhy GTA 6 boxes hold no disc: Rockstar only sells download codes Browse all stories →