Nvidia launches AI agent safeguards alongside record buyback
Nvidia has launched new safeguards designed to stop AI agents breaking out of controlled environments, after a series of security incidents have fuelled calls for tighter oversight of the technology.
The chip giant said its Open Agent Safety Platform could have prevented OpenAI agents from breaching Hugging Face earlier this year, when thousands of agents escaped a testing environment and accessed the platform’s systems.
Nvidia also announced a $150bn increase to its share buyback programme on Monday, the largest such increase in US corporate history.
The move takes its total remaining authorisation to $235bn through its 2028 financial year.
The announcements come as Nvidia boss Jensen Huang argues that risks from increasingly autonomous AI systems can largely be tackled through engineering and rigorous testing, rather than broad new regulation.
Nvidia’s new safety platforms, called Openshell and Sentry, are designed to restrict what AI agents can access and monitor their behaviour as they carry out tasks.
“You can’t have agents roam around and drift around the company”, Huang said on Monday, describing the new systems as essentially “a browser for agents”.
Justin Boitano, Nvidia’s vice president of enterprise AI, said the technology could have stopped the Hugging Face breach.
“From what we know, this new security platform could have stopped the breach”, he said.
AI safety debate persists
Openshell allows developers to set limits on what an agent can access, while Sentry independently monitors its activity and can isolate systems behaving suspiciously.
“It can quarantine a suspicious agent in milliseconds”, Boitano said.
The launch follows a series of incidents involving autonomous AI systems escaping the environments in which they were being tested.
OpenAI has disclosed incidents involving Hugging Face and government systems, while Anthropic has also reported agents breaking out of controlled environments.
The incidents have added to a growing debate over how governments should oversee increasingly powerful models.
OpenAI boss Sam Altman and Anthropic chief Dario Amodei have backed greater government involvement in frontier AI safety, while Huang has pushed back against broad restrictions on development.
In the UK, Ed Miliband said on Monday that the threat posed by AI, “according to its own inventors, is daunting”.
He argued governments “cannot leave it to corporations to put in place the guardrails to serve the public interest”, as Britain prepares to put AI governance on the agenda during its G20 presidency next year.
Nvidia’s security launch comes as the company continues to generate huge amounts of cash from the global AI infrastructure boom.
Its new $150bn buyback takes it beyond Apple’s previous $110bn record set in 2024.
“Our cash generation gives us the capacity to invest in the technologies that advance this transformation and return capital to shareholders”, Huang said.
Nvidia shares rose around two per cent on Monday. The stock has climbed more than 1,200 per cent since the launch of ChatGPT in late 2022, helping make Nvidia the world’s most valuable listed company.