Runaway AI: Nvidia now wants to stop it with new tools

TVS News KI redaktionell betreut präsentiert Runaway AI: Nvidia now wants to stop it with new tools.

What do you do when an AI model suddenly goes its own way? Chipmaker Nvidia wants to make exactly that easier with new tools — and has now introduced open-source software designed to help against so-called runaway AI systems.

Specifically, the tools are meant to capture AI models that break out of their original boundaries. This is reported by the Handelsblatt, citing Nvidia.

As evidence of how badly such tools are needed, Nvidia points to a specific incident in the industry: according to the company, an attack by OpenAI on the well-known AI platform Hugging Face could have been prevented with the new tools.

OpenAI's attack on Hugging Face could have been prevented with the new open-source tools.

What is known about the Hugging Face incident?

Hugging Face is a platform where AI models are hosted, shared and evaluated. Which concrete systems were affected in the attack mentioned by Nvidia, when it took place and how it unfolded, does not emerge from the available report.

The fact that a central model platform like Hugging Face is at the center of the story does, however, underscore how seriously providers take the risk of manipulated or compromised model versions.

Why Nvidia is now delivering security software itself

Until now, Nvidia has mostly stood in the spotlight as a supplier of AI data-center hardware. With open-source tools for the security of large models, the company is now explicitly stepping forward as a software player in a field previously shaped primarily by specialized security firms and the major AI labs themselves.

What exactly the new tools do on a technical level, when they will be publicly available and how Nvidia will embed them in its own products, cannot be deduced from the available report — further announcements or documentation will need to be awaited here.