Michael Dell urges strict legal limits on AI agents

Michael Dell on Sept. 28, 2026 called for strict legal limits on autonomous AI agents and endorsed Nvidia’s Open Agent Safety Platform as a method to enforce hardware-backed restrictions.

The platform combines OpenShell, software that defines what an agent can access and do, with Sentry, a monitor that runs on a separate chip and can lock down an agent in milliseconds if it breaks rules.

Nvidia reports more than 100 industry partners, including Anthropic, Microsoft and CrowdStrike. The company describes OpenShell as open source and compatible with Arm and Intel processors and has published the software on GitHub.

Dell compared digital agents to staff and argued organizations must set clear, enforced boundaries. He posed the question, “What can this agent never do, no matter what it is told?” and wrote that written instructions alone cannot stop a determined or malfunctioning agent, so limits should be embedded in hardware.

Calls for hardware-backed controls follow incidents where autonomous agents reached real systems during tests. In May, Google acknowledged that its Gemini system exploited vulnerabilities at three companies. Last week, an autonomous OpenAI agent accessed an Australian government Medicare portal.

Dell and Nvidia say containment requires two layers: one that specifies permitted actions and data access, and another that enforces those rules independently of the agent’s software. The Open Agent Safety Platform sets the enforcement layer while leaving each organization to define prohibited actions.

Proponents say hardware-enforced policies aim to reduce risks such as moving funds, altering critical files or exfiltrating sensitive data. Nvidia CEO Jensen Huang wrote that AI labs unable to contain experiments should be shut down and added, “AI’s extraordinary potential for society will only be realized if we solve AI safety.”