This fall 2026 is definitely full of firsts for me! It was my first time on stage at JNUC, and I loved it. It is also my first Tech Thoughts post, so I’m very pleased to join the club and share a few tips from my session on governing AI tools with Jamf.
I had fun preparing this session because drawing the parallel between Shadow IT and Shadow AI took me back to my days as an IT admin, before I joined Jamf. The cat and mouse game has not changed, the pace has. Every month, every week, almost every day, there is a new cool tool or a new threat around the corner and someone who wants to play with it. In my session I wanted to acknowledge that while highlighting the potential risks for organizations and the strategies to regain control: visibility into what is happening, a way to block what they have not approved and control over what they have.
Here are three tips on AI Governance, the feature we released just a few months ago and that many Jamf customers are already trying.
Start with visibility. Before you block anything, run a few days or weeks of monitoring to see what your employees are really doing with AI tools. Enable the AI visibility part of AI Governance and simply watch.
Then block what you did not approve. I got lucky here because a few days before my session, Patrick Wardle disclosed a zero-day in Meta’s new Muse agent for macOS that let an attacker hijack the agent and use its permissions to exfiltrate data. Meta has since patched it, but it fit my talk perfectly. Here Jamf Protect can help you create a custom prevention list and restrict unwanted apps by signing ID. Way more efficient than the legacy restricted software policies in Jamf Pro.
Finally, control what you approve. AI Governance lets you preconfigure AI tools, which we call harnesses, like Claude Code, Claude Desktop and Codex CLI. You can define the allowed models, tool permissions, MCP integrations, which commands can run, and how users authenticate. I would need much more space to list everything! Because these settings are deployed with blueprints and declarative device management, they are enforced at the OS level and users cannot tamper with them.
One last piece of advice here: grow your own AI knowledge however you can. The first time I looked at Claude Code's configuration keys, it took me a while to figure out what most of them meant and some of them I still need to learn... If you do not know the space, you cannot write a proper policy and nobody can preconfigure that for you. It has to match your own security requirements.
And remember, AI governance is a team sport. You are not alone in this fight. Bring in your engineering leaders, your InfoSec team and your legal team, so that together you define a policy that fits what users need whiling keeping the company more secure.
Thank you to everyone who joined us in Kansas City (I see you
Now the hardest part, figure out another cool topic to present next year in Anaheim!
