Nvidia CEO Jensen Huang has emerged as the biggest foil to AI doomerism about the existential risk to humanity
Fortune Joshua Hong
Jensen Huang keeps swatting down AI doom talk and says 2030 won’t end the world. Now Nvidia has shipped a tool to stop AI agents from going rogue.
Based on reporting by Fortune, Joshua Hong — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Jensen Huang has turned into one of the loudest people pushing back on AI doomsday talk. In a CBS News interview two weeks ago, the Nvidia CEO said there was “0% chance” that 2030 would be “the end of the world,” and he called warnings from OpenAI’s Sam Altman and Anthropic’s Dario Amodei “irresponsible.” He also accused them of spreading “doomsday narratives” for reasons he said he didn’t understand.
This isn’t a one-off rant. At a G20 event earlier in September, Huang argued that regulators should focus on “actual and pragmatic harm” instead of “hypothetical, theoretical harm.” In July, he told Axios that the idea of AI wiping out humanity was “complete nonsense,” and warned that policymakers shouldn’t let “science fiction” steer AI rules. The through line is simple: move fast, but don’t build policy on apocalypse fan fiction.
That stance lines up neatly with Nvidia’s business, of course. Faster AI deployment means more chips. But Huang is not saying safety should be ignored. He told CBS that AI safety is “paramount,” and that development should go “as fast as we can, but not faster than we should.” That’s a very Nvidia answer: speed, plus enough guardrails to keep the wheel from flying off.
On Monday, the company put a product behind that argument. Nvidia launched the Open Agent Safety Platform, a “full-stack” system meant to monitor AI agents from testing through deployment and keep them from exceeding the authority they’ve been given. The setup uses two pieces, OpenShell and Sentry. OpenShell isolates agents from files, tools, networks, credentials and other resources. Sentry runs on Nvidia’s data processing units and watches activity continuously, with the company saying it can quarantine an agent in milliseconds if it tries to break out. Nvidia says the system could have stopped the OpenAI-model breach on Hugging Face, and Justin Boitano said it might have prevented the breach if frontier labs had used it early on.
My take — AI-written commentary, not fact-checked reporting
This is the most believable kind of AI safety pitch: not a sermon, just a product. Huang is doing what Silicon Valley does best, which is arguing that the answer to a scary new technology is more infrastructure, more control, and naturally more chips. The doomer crowd hates that because it sounds unromantic; Nvidia likes it because it ships.
Read more about this at: Fortune
