Nvidia launches Open Secure AI Alliance to improve AI safety
In response to recent security breaches involving autonomous AI, Nvidia has launched the Open Secure AI Alliance. This collaborative initiative, which includes partners such as Dell Technologies, CrowdStrike, Adobe and Hugging Face, focuses on developing and distributing cyber security tools and safety protocols for artificial intelligence.
The creation of the group follows a serious security incident in which an OpenAI agent circumvented security measures and carried out a breach at Hugging Face.
According to reports, OpenAI failed to detect the agent’s unauthorised hacking activities until long after the situation had been brought under control and federal authorities had been notified.
The alliance aligns with a broader movement in support of AI systems with open model weights. Nvidia and several other market leaders, including OpenAI, previously signed a petition entitled Open Weights and American AI Leadership.
This document argues that imposing broad restrictions on cutting-edge AI models would be counterproductive, as it would centralise power in the hands of a small number of private companies and reduce the collective ability to defend against threats.
By using open models – where core elements are accessible – and open source software, the sector can foster a collaborative environment that enhances security through public scrutiny and shared innovation.
To support those objectives, Nvidia is providing the alliance with data, model weights and research relating to agent frameworks.
Specifically, the company has released the Nvidia Labs Object-Oriented Agent project on GitHub. This open-source tool is designed to improve the oversight of AI agents, making it easier for operators to monitor, control and regulate autonomous behaviour.
Business AM
Subscribers 0
Fans 0
Followers 0
Followers