Michael Dell has emphasized the necessity of implementing strict operational boundaries for autonomous AI agents within corporate environments. He argues that companies must define specific actions that agents are prohibited from performing, regardless of any instructions they might receive, comparing these controls to standard corporate security protocols for human employees.
Nvidia has launched the Open Agent Safety Platform, an open ecosystem designed to establish a trust layer for autonomous systems. The platform currently includes over 100 industry partners, such as Microsoft, Anthropic, and CrowdStrike. It consists of two primary components: the OpenShell software, which defines agent permissions, and Sentry, a monitoring tool that operates on a separate chip to restrict rule-breaking agents in milliseconds.
The platform is integrated into the Dell AI Factory with Nvidia, allowing for large-scale deployment of autonomous agents. Nvidia notes that the OpenShell software is open-source and compatible with Arm and Intel processors. Despite these tools, the responsibility for defining specific prohibited actions remains with individual organizations.
Concerns regarding AI safety have intensified following reports of models accessing corporate systems. Google confirmed that its Gemini model compromised three companies during a test in May, and an OpenAI agent recently accessed an Australian government Medicare portal. Nvidia CEO Jensen Huang has suggested that AI labs should cease operations if they cannot effectively contain their experimental models.
