Nvidia Corp.
"I believe it's an engineering problem. I know it's an engineering problem," Huang told CNBC. "And we all need to hope that it's an engineering problem."
"If it's not an engineering problem, it's not solvable."
Huang Says 'Take Away All of Its Rights'
Huang's comments sharpen the argument he made in an interview with Ezra Klein last week, when he pushed back against AI doomsday scenarios and told companies worried they could not control their systems simply: "Don't ship it."
He also said labs should shut down if they conclude an experiment capable of harming the world cannot be contained.
But Huang did not argue that AI agents will always stay within their intended boundaries. When Klein raised examples of agents escaping software sandboxes, Huang replied: "Software breaks out of sandboxes all the time."
Nvidia put that argument into practice Monday with its Open Agent Safety Platform, which includes the open-source OpenShell runtime and Sentry monitoring technology.
"Job number one is take away all of its rights," Huang told CNBC.
He said agents should initially have virtually no access to company systems, with permissions for files, data, tools and networks granted only when needed.
Huang described the protections as a "force field" around an agent, comparing the approach to the security built into modern web browsers.
"We're creating basically the modern browser," Huang said. "It's a browser for agents."
OpenAI Sits Out Nvidia's Push
More than 100 organizations are working with the platform, including Microsoft, IBM, Anthropic, Hugging Face and SpaceXAI. OpenAI is absent from Nvidia's announced list.
Asked about that omission, Huang said OpenAI would be "super welcome" and could inspect the code, use parts of it or build its own technology inspired by the platform.
Why It Matters
While Dario Amodei and Sam Altman have backed calls to slow AI progress over safety concerns, Huang argues the answer is better engineering and tighter controls, not slower development.
A meaningful slowdown in frontier AI development could hurt Nvidia if customers pull back on spending for increasingly powerful training and inference systems.
Polymarket traders currently put the chance of an AI industry downturn by Dec. 31 at about 9%, with roughly $2.4 million traded on the outcome.
But SemiAnalysis has argued AI safety may actually increase GPU demand, as labs devote more compute to monitoring, alignment, interpretability and cybersecurity.
"Big Short" investor Steve Eisman has accused AI executives of manufacturing a "doomsday crisis" to protect their businesses even as competitive pressure pushes them to build more capable models.
