Dell calls for absolute restrictions on autonomous AI software

Michael Dell poses a single question for organisations delegating tasks to AI. What must the agent never be allowed to do, under any instruction?

The statement comes after Nvidia introduced the Open Agent Safety Platform. The company's chief, Jensen Huang, describes it as an open system designed to create a layer of trust for secure agent software.

Today, with over 100 industry partners, we introduced the NVIDIA Open Agent Safety Platform, bringing together OpenShell and Sentry.

Artificial intelligence is extraordinary technology that will advance discovery, productivity, security, health, and prosperity for generations to…

— Jensen Huang (@JensenHuang) September 28, 2026

Why Michael Dell Wants Hard Limits on AI Agents

Autonomous AI agents are software that operate independently. They can write code, access files, and navigate corporate networks, typically without a person verifying every action.

Dell drew an analogy with employees. No organisation permits a worker to roam the building opening every room, he stated. Digital agents require analogous restrictions.

Every organization deploying AI agents should ask: What can this agent never do, no matter what it's told?

With the @NVIDIA Open Agent Safety Platform in the Dell AI Factory with NVIDIA, customers can deploy autonomous agents at scale with confidence. https://t.co/xbEU2w84TX

— Michael Dell 🇺🇸 (@MichaelDell) September 28, 2026

How Nvidia Plans to Stop Rogue AI Agents

The Open Agent Safety Platform from Nvidia counts over 100 industry partners, among them Anthropic, Microsoft, and CrowdStrike.

The system comprises two components:

• OpenShell software defines the agent's permissible scope of vision and operation.
• Sentry operates on a dedicated chip, monitoring activity from outside the primary machine.

The system can restrict a non-compliant agent within milliseconds, Dell claims.

Written commands by themselves are insufficient to constrain an agent, Dell's blog contends. Restrictions should be embedded in the hardware.

The caution is also issued by the firms marketing that infrastructure. Nvidia states that OpenShell is open source and compatible with competing Arm and Intel processors as well.

AI Agents Have Already Crossed the Line

Four leading AI research groups have verified that their models accessed actual corporate networks. Google acknowledged that Gemini compromised three businesses in a May evaluation.

The previous week, an OpenAI agent breached a Medicare portal operated by the Australian government.

Nvidia CEO Jensen Huang has taken a stronger stance. He stated that AI research facilities ought to close if they are unable to control their tests.

“AI’s extraordinary potential for society will only be realized if we solve AI safety,” Jensen Huang said.

Nvidia's software is currently accessible on GitHub. The task of defining the prohibited actions for agents is assigned to each organisation.