• NVIDIA (NVDA) CEO Jensen Huang reaffirms commitment to rapid AI advancement with robust safety measures, as the company launches its Open Agent Safety Platform.
  • The platform, introduced on September 28, aims to keep autonomous AI agents within defined permissions using open-source software and hardware-backed monitoring.
  • Huang opposes blanket AI regulation but supports tailored rules for high-risk applications, balancing innovation with accountability.

NVIDIA's Dual Push: Speed and Safety

In the race to deploy increasingly autonomous AI agents, NVIDIA CEO Jensen Huang is betting that safety and speed can coexist. "We're going to advance this responsibly and safely," Huang has said, a message underscored by the company's September 28 launch of its Open Agent Safety Platform. The platform, which combines open-source runtime software with hardware-backed monitoring, aims to prevent AI agents from overstepping their bounds—a growing concern as businesses deploy agents that can execute multistep actions across enterprise systems.

The launch comes amid rising anxiety about long-running AI agents bypassing application-layer controls, potentially leading to unauthorized access, data leakage, or unsafe physical actions. NVIDIA's answer is a full-stack approach: controls embedded in software, compute infrastructure, and hardware. At its core is OpenShell, an open-source runtime that records agent actions and restricts what an agent can access and do. Paired with Sentry, a watchdog running on NVIDIA's BlueField-4 data-processing units, the system can monitor and quarantine an agent within milliseconds if it attempts to exceed its assigned boundary.

Huang's stance is nuanced. He argues that companies should not release systems they cannot properly test or control, yet he opposes blanket AI-safety regulation and rejects special exemptions from existing antitrust or product-liability rules. Instead, he supports tailored regulation for high-risk deployed products, such as robotaxis. This middle-ground approach contrasts with calls from some AI labs for more formal pre-deployment rules and external audits. "Institutional investors are really focused on regulatory stability," Huang has said in other contexts, and his position reflects a desire to avoid a patchwork of rules that could stifle innovation.

The platform has already garnered support from over 100 organizations, including Anthropic, Cisco (CSCO), Microsoft (MSFT), SAP (SAP), Salesforce (CRM), JPMorganChase (JPM), Palo Alto Networks (PANW), CrowdStrike (CRWD), Perplexity, and Red Hat. Their involvement signals a broad industry appetite for practical safety tools that don't require sweeping new laws. For enterprises, the platform offers auditability and containment when agents touch sensitive data or operational systems. For developers, it provides open-source building blocks, though implementation quality will remain critical.

Financially, NVIDIA is riding an AI infrastructure boom. In its fourth quarter of fiscal 2026, revenue hit $68.1 billion, up 20% sequentially and 73% year over year, with net income doubling to $43.0 billion. Data Center revenue alone was $62.3 billion in the quarter and $193.7 billion for the year. The company returned $41.1 billion to shareholders through repurchases and dividends and ended the quarter with $58.5 billion remaining under its buyback authorization. Looking ahead, NVIDIA forecasts first-quarter fiscal 2027 revenue of $78.0 billion, plus or minus 2%, though that outlook assumes no Data Center compute revenue from China—a reflection of U.S. export controls that have effectively shut NVIDIA out of the Chinese market.

The new safety platform could deepen reliance on NVIDIA's hardware and software stack, expanding its competitive moat from AI compute into the governance and security layer of agent deployment. It also addresses a major adoption obstacle: security and liability risk. If widely adopted, runtime containment and independently enforced access controls could become baseline requirements for enterprise agents, much like identity management and endpoint protection did for cybersecurity.

Still, questions remain. NVIDIA's descriptions are product claims, and real-world effectiveness will need validation through independent security testing and deployments under adversarial conditions. The company itself cautions that many capabilities remain subject to development timing and other risks. Moreover, voluntary technical safeguards may not satisfy regulators who argue that mandatory reporting, legal liability, and external oversight are necessary. Civil society groups may welcome concrete tools but question whether they are enough without enforcement.

Huang's message is clear: he is not calling for slower AI development. He is advocating fast development coupled with testing, product accountability, and engineered containment. As AI agents move from generating content to executing complex tasks, NVIDIA is positioning itself as both the engine and the brake—a strategy that could define the next phase of the AI revolution. The company's financial strength gives it ample resources to pursue this vision, but the ultimate test will be whether its safety tools can withstand the ingenuity of those who seek to bypass them. NVIDIA did not respond to a request for comment on the platform's independent testing protocols.

Correction: An earlier version of this article misstated the quarter in which NVIDIA will begin including stock-based compensation in non-GAAP measures. It is Q1 FY2027, not Q4 FY2026.