AI agent demos look impressive. Getting one past your IT department is another story. Agents can connect to your systems freely and leave no record of what they touched. For enterprises running sensitive data across dozens of connected systems, that’s not a risk most organisations can justify.
That’s changing. Last week, NVIDIA founder and CEO Jensen Huang joined ServiceNow chairman and CEO Bill McDermott on stage at Knowledge 2026 to announce a partnership that targets exactly that problem: how to give agents real autonomy across workflows without losing visibility or control.
The partnership introduces Project Arc, an AI agent that can work across your files and business systems, but only within rules your IT team sets in advance. Think of it as an agent with a defined job description it cannot go outside of. NVIDIA powers the AI; ServiceNow provides the business workflows and controls it operates within.
Why is Project Arc considered safer than other AI agents?
Two things set it apart from most agents on the market, according to NVIDIA’s announcement. First, every action Project Arc takes is logged through ServiceNow’s AI Control Tower, recording what it did, when, and who authorised it.
Second, a tool called OpenShell controls what the agent can actually access, letting IT define hard boundaries before it touches anything live. Most agents offer neither. That combination of a full audit trail and enforced access limits is what makes Project Arc viable in environments where a single unchecked action could have real consequences.
That matters because governance isn’t just a technical requirement. It’s what turns a pilot into a real deployment. Enterprises have been testing AI agents for years. The ones that haven’t scaled aren’t waiting on better AI. They’re waiting on something IT and legal teams can actually sign off on with confidence.
NVIDIA’s role in the partnership
Most people know NVIDIA for its chips, but this partnership goes beyond hardware. NVIDIA is contributing pre-built AI capabilities that businesses can plug directly into their workflows, a toolkit for developers building on the platform, and NOWAI-Bench, a new way to test and compare how enterprise AI agents actually perform in the real world.
NVIDIA’s own Nemotron 3 Super is, according to the company, currently the top-ranked open-source model on that benchmark. The company has also made running agents at scale more cost-efficient, meaning businesses can keep agents working continuously without it becoming too expensive to justify.

What Australian businesses need to know about AI agents
This announcement came at an interesting moment for Australia. In the past two weeks, three local regulatory bodies — the Australian Signals Directorate, ACSC and APRA — all released separate guidance flagging AI agents as a specific risk. Businesses, they warned, are deploying agents faster than they’re putting controls in place.
This is where the NVIDIA-ServiceNow partnership comes in. An AI agent that records its actions and stays within set limits is a much easier sell to Australian organisations than one that operates freely. It’s one of the first things that comes up when we work with businesses on automation at AIVY, and it’s rarely about the technology.
It’s about getting it approved. Knowing what an agent can access, and having a record of what it did, is often the difference between a project going live and staying in a pilot. It’s exactly the kind of problem we help businesses work through.
What this means for the future of AI agents
When two companies the size of NVIDIA and ServiceNow make governance their central product argument, it sets a new standard for the whole industry. Other AI agent platforms will have to respond. For Australian businesses that have been watching from the sidelines, this is a signal worth paying attention to. The experimental phase of enterprise AI agents is winding down.
The expectation now is governance built in from the start. Not as a future upgrade, but as a baseline. The businesses that figure that out first will have a meaningful head start. The first question worth asking isn’t which agent to deploy. It’s whether your governance framework is ready for one.
