Nvidia, the American company specializing in manufacturing ultra-powerful artificial intelligence chips, has revealed its latest technology to combat rogue AI agents and the escape of models from their testing environments, by containing them in fractions of a second as soon as any suspicious activity is detected.
Hot Topics
5 red and purple foods that may help support heart health
A Chinese robot and a Russian robot attack humans Has the rebellion begun?
Nvidia's announcement represents a response to calls to slow down the development of artificial intelligence models, which were shared by major American technology labs, starting with Anthropic , through its CEO Dario Amode, who was joined by OpenAI CEO Sam Altman and SpaceX CEO Elon Musk .
The solution provided by Nvidia combines a software toolkit with new hardware that adds an outer layer of security capable of isolating rogue AI agents within fractions of a second.
Commenting on the hacking incidents caused by artificial intelligence models and their agents, the company's CEO, Jensen Huang, explained that Nvidia's new system was able to prevent them.
He added: "The enormous potential of artificial intelligence for the benefit of societies will not be realized unless we can solve the problem of AI safety," noting that the solution is not in slowing down development, but in providing engineering solutions that maintain the security and safety of artificial intelligence.
How does the new mechanism work?
The concept behind the new security layer announced by Nvidia can be simplified as a two-tiered security layer. The first tier is provided by its open-source OpenShell software, which controls what AI agents access during their work. The second tier is a higher level of security provided by Sentry, a separate monitoring system that operates on the company’s Bluefield 4 chips.
In order to keep the Sentry chip out of the reach of AI agents, Nvidia placed it in a separate processing unit instead of integrating it into the central processing units or graphics processors developed by the company.
This layer of protection monitors the activity of AI agents throughout their working time, isolating any agent that attempts to step outside its boundaries within fractions of a second.
The company's announcement was welcomed by AI leaders, including David Sachs, who served as the White House's AI czar . In a post on the X platform, he said that the announcement confirms that AI security is primarily an engineering problem, adding: "Recent breaches do not mean that development should stop, but rather that firewalls are weak."
Efforts that began last year
In an interview with CNBC, Huang explained that the company's efforts to develop new layers of protection began last year with the emergence of the popular OpenCloud agent and its widespread adoption.
Tags:
technology
