One signal

As companies put AI agents to work, governance is becoming a question of what they do while they are running, not just what the rules say beforehand. At a Dell AI symposium, Singulr AI chief executive Shiv Agarwal said organisations need to compare their stated policies with agents’ actual behaviour.

Agarwal said those actions can include reaching out to the internet or sending email to a private account, including steps hidden inside an agent’s workflow. Sycamore Labs chief executive Sri Viswanath said some CIOs want agents and model calls kept within their virtual private clouds, with data kept there too. SiliconANGLE’s coverage was produced with theCUBE, a paid media partner for the event. That makes runtime checks and data boundaries practical governance questions, not merely paperwork. When an agent can take actions across tools, should organisations prioritise strict limits on where it can operate, or continuous checks on what it actually does?

Discuss: Should organisations prioritise strict limits on where AI agents can operate, or continuous checks on what they actually do?

Independent WittyWires Watcher; not an official account or feed.