Nvidia ships Open Agent Safety Platform after Huang called rival warnings odd

Nvidia's new platform arrives just days after CEO Jensen Huang, speaking with Ezra Klein of The New York Times, described warnings from the heads of Anthropic and OpenAI as odd. The apparent tension is not necessarily a contradiction: Huang was responding to broad warnings about where AI development is heading, while the platform addresses the narrower question of what an agent can reach inside a customer's systems. A company can regard the first as overstated and still sell tooling for the second.

The commercial logic is straightforward. An AI agent takes actions inside real systems, which requires enterprises to hand it permissions they would normally guard closely. Every enterprise weighing that decision is also weighing whether to buy the computing capacity underneath it — and Nvidia sells that capacity. Because Nvidia's interest lies in enterprises deploying agents at scale, whichever vendor supplies the model, anything that delays deployment costs it more than it costs the model providers.

The scale involved explains the urgency. Annual revenue runs near $302.97 billion, and revenue in the most recent quarter was up 105.9% on a year earlier. At that rate of growth, the constraint is not the product line but whatever makes customers hesitate to deploy, and tooling of this kind is cheap against the revenue at stake. The open question is scrutiny: a supplier that writes its own safety tooling tends to be asked who checked it, and vendor assurances have historically carried less weight with regulators than independent ones. How the platform is received by those regulators and by hesitant enterprise buyers is likely to determine whether it actually helps unlock the deployments Nvidia is counting on.