On 20 August we were at the AI & Data Governance Summit 2026 (ADG Summit) at Rydges Melbourne, run by DGX Group. We came as a gold sponsor, with a table in the sponsor hall, a TV running our product demo on a loop, and a 25-minute demo session over lunch.
DGX told us 239 people registered and 206 checked in on the day. DGX describes ADG as an event built by practitioners for practitioners, and the agenda held to that. Most sessions were people running governance programs inside real organisations, talking about what works and what doesn't.
Conference chair Chad Barendse set the tone in the opening: AI has not created the governance problem, it has exposed gaps that were already there.
The panel on practical AI governance
Paul Towers, our sales manager for Australia and New Zealand, moderated the 12:00 panel, "Practical AI Governance: What It Actually Takes". On stage with him were Fernando Mourão (Head of Responsible AI, SEEK), Paola Oliva Altamirano (Director of Data, AI and Analytics, SmartyGrants) and Bikram Choudhury (Responsible AI, Accenture).
Paul opened with something most of the room already knew: "There's no shortage of AI principles, frameworks, and policy documents." What people are short on, he said, is clarity on "what AI governance actually looks like in a day-to-day world." The points that stuck with us:
- Policies and frameworks are the skeleton. Fernando's view was that governance is what people do every day, and the document on its own isn't it.
- Bikram described governance as being able to answer a handful of questions: what AI you have, where it sits, who owns it, who is accountable for it, and whether it should have access to personal data.
- Not every use case needs the same process. Low-risk work can be fast-tracked so the effort goes into the high-risk cases.
- Start with the problem, not the tool. When SmartyGrants put a data scientist on call for staff, Paola said a lot of the friction was fixed by looking at the workflow, sometimes with nothing more than a dashboard or a small app. New tools get tried on mock data, never client data, before they go through a risk review.
- When picking tools, Fernando looks for flexibility and a lean start, and avoids anything that locks you in later.
- Start governance early. Fernando said it speeds delivery when it is there from the beginning and slows things down when it is bolted on later. He also said he spends about seventy percent of his time talking to people.
On agents, Fernando argued for designing for testability: the first agents you build should be the ones that test your product agents. Paola suggested treating a new agent like a new hire, with a defined role, a probation period and regular reviews.
The last audience question asked how to use AI to keep up with the speed of change. Bikram said organisations are already using AI to govern AI, with a human in the lead. Paul closed the session on that line: "AI governing AI."
What we took away
Most of what the panel described is about people and process, and no platform fixes that for you. Some of it does depend on the layer underneath, though. If you want to know which models are running, who deployed them and who is allowed to change them, the platform has to record it.
That is the part Xinference covers. Teams deploy open models on infrastructure they choose and call them through one API, with role-based access control, audit logs and OIDC single sign-on. It won't write your AI policy. It does make several of Bikram's questions easier to answer.
Thank you
Thanks to Tim Moon, Chad Barendse and the DGX Group team for a well-run day, and for the photos in this post. Thanks also to Fernando, Paola and Bikram for their time and their candour on stage.
If your team is working through the same questions, book a demo. And if control over your AI stack is on your mind, join our free webinar with BioQuest Advisory on 15 October, Who controls your enterprise AI?