Home Archives for Dan Morgan
Summary
The Australian Federal Government’s new National AI Centre is moving from voluntary guidance to a more prescriptive framework indicating that formal legislation may not be far off. Five areas are now directly in the sights of Parliament: duty of care, privacy, workplace safety, consumer protection, and automated decision making. Businesses that build governance now avoid an expensive scramble later.
What’s Changing in Australian AI Regulation
For two years, Australian businesses treated AI governance as optional. The Federal Government took a relatively “hands-off” approach to AI regulation, the extent of regulation contained within the Voluntary AI Safety Standard1, with no penalty for skipping it. This has now changed.
In August 2026, the Department of Industry, Science and Resources signalled a shift toward a standards-led, enforceable framework2 for artificial intelligence. The NAIC has now published formal guidance on the following topics:
- Duty of Care
- Privacy
- Workplace Safety
- Consumer Protection
- Automated Decision Making
The announcement follows a rapid build-up4 of legal cases5 involving AI misuse6. Yet the government is walking a careful line between regulation and growth.
Risk, compliance, and legal teams have first-hand knowledge of how quickly guidance is changing and how difficult it can be to form a defensible AI-use policy.
During Dell Tech Forum 2026 in Sydney, senior industry leaders spoke about their experiences in forming internal operating procedures in the absence of official guidance, finding that adopting a core set of “common sense” principles based on recent EU AI Act has achieved the most success in satisfying the risk appetite of boardrooms.
While the absence of specific law may be an intentional move from Parliament as “legislating for every conceivable risk would only push investment offshore”7), there are steps Australian businesses can take to today to reduce the chance of ending up in a precedent-setting legal case.
The “Essential Six” You Need to Implement
Accountability
One or more people inside your business now must be held accountable for the actions of AI and agents. AI complexity can create gaps where no one takes clear responsibility for outcomes, especially as organisations move further toward autonomous agents. Without accountability, AI and agents are now able to materially affect business trade and reputation with no redress.
Impact & Disaster Recovery
Disaster recovery in IT used to be limited to infrastructure failure, then expanding to cyber events. Both were considered unlikely and anomalous situations. With the increased reliance on AI and agents, and the speed in which they act, negative actions within a business can spread incredibly fast.
Organisations now need to have a plan for recovery when AI and agents fill your Salesforce with incorrect data, delete Sharepoint repositories, and any other destructive actions that agents may make.
Maintain an AI Register
With AI becoming more autonomous, it is now critical to assign the same level of identity, record-keeping, and information to them as we do with hired human staff. Each AI or agent should have an accountable individual, organisational identity (for authorisation and auditing), impact & risk assessments, and be transparent in its fallibility.
Don’t Patch – Monitor
The main source of change-related pain in IT used to be the Russian roulette of bugs introduced with scheduled patching. With the advent of agents, this has changed to unintended behaviour change as new models are released, bias within training data or post-training activities. This is especially important in industries particular conscious of AI sovereignty.
Organisations must be able to:
- Generate a baseline of expected behaviour and the general desired outcome or quality of product of an LLM and
- Be able to monitor changes to this baseline or quality and
- Act accordingly
This was also a hot topic during Dell Tech Forum 2026, with organisations advanced in their AI adoption speaking about the importance of configuring and maintaining telemetry monitoring of agents and their output.
Kill Switch
LLMs are non-deterministic – unlike traditional software that follows explicit instructions, AI systems learn patterns from data and make their own opaque decision logic. This means unintended output may cause material damage to business.
Organisations must ensure individual/s overseeing your AI systems should know how to do so appropriately, and what they need to do to override the system if something goes wrong.
Where Australian Businesses Stand Today
We talk to organisations across every stage of AI maturity, and two clear cohorts have emerged.
Early-Stage Organisations (80%)
Around 80% of the organisations we speak with are early in their AI journey. Most run off-the-shelf co–work platforms, isolated pilots, or point tools rather than a governed platform with centralised logging and access control.
If that sounds like your business, you’re actually well positioned to design a data and AI architecture from the ground up, guided by the National AI Centre’s foundation recommendations7, instead of retrofitting governance onto production systems already carrying technical debt.
If you are in this cohort and have AI aspirations, we recommend prioritising and building out a data strategy and platform as soon as possible. Our data shows that, for organisations that do not have a formalised data strategy and platform, 95% of AI projects fail to produce any business value.
Advanced Organisations (20%)
The remaining 20% have an established, mature, and heavily used data platform, are fostering citizen developers, and experimenting with autonomous agents. If this describes you, this announcement should shape your AI governance charter directly, particularly around model monitoring and decision auditability. The NAIC also publishes implementation guidance8 for organisations operating at this level of complexity.
If you are in this cohort, you’re likely beginning to struggle with managing the breadth of your data estate and might even be running into “tokenomics hell” – frontier or cloud LLMs causing costs to overrun.
Data Sovereignty is Now AI Sovereignty
Data sovereignty used to describe something fairly contained: keeping data within Australian borders, subject to Australian law. AI has stretched that definition well beyond storage or inferencing location.
AI sovereignty spans the full technical stack: where inferencing executes, which jurisdiction holds the encryption keys, who controls the model weights, who has access to the MCP servers, and who owns the energy infrastructure powering the data centre.
In practice, AI sovereignty means your organisation can verify every layer beneath an AI system, from model to physical site, remains under domestic control. That includes resilience against disruption at the infrastructure layer, not just the application layer.
Government will legislate the national dimension of this. But organisational AI sovereignty is increasingly a procurement requirement. A year ago, around 9% of organisations we spoke with ranked a sovereign AI platform among their top priorities. Within six months, that figure jumped to 33%, concentrated in education, research, and healthcare, sectors with strict data classification requirements.
The Australian Signals Directorate’s Cyber Security Centre reinforces this expectation directly. It advises organisations to confirm that any AI system they use can meet their data residency and sovereignty obligations9.
Our Sovereign AI Journey
As a proud Australian technology company, we wanted to become experts at this. We rebuilt our data foundations on Microsoft Fabric, in-housed our inferencing, and migrated all our agents into our own sovereign datacentres. We structured information into governed bronze, silver, and gold layers, with lineage and access policy enforced before any dataset reached a model. That structure now underpins our business decisions and every customer interaction via our home-grown Customer Insights platform10.
In our first-hand experience, we can say confidently that clean, curated, and centralised data is the architectural backbone the entire system depends on.
Two Technical Paths to Compliant AI
Extend Your Existing Microsoft Investment
If you’ve already invested in Microsoft, the shortest path to return on investment is with Copilot, building data agents natively within Fabric using its existing governance and access controls. Either path builds on infrastructure you likely already licence.
Deploy a Sovereign Platform
If sovereignty is the priority, Launch AI is worth evaluating. It’s backed by the technology group trusted by 42% of Federal Government, and it delivers predictable pricing alongside enterprise-grade governance. It gives you control over data residency at rest and in transit, customer-managed encryption keys, and defined inferencing regions, built on ISO 27001 and PCI-DSS certified infrastructure.
Neither path requires starting from scratch. Both require starting now, while the runway before legislation remains generous.
Frequently asked questions
What is the difference between data sovereignty and AI sovereignty?
Data sovereignty covers where data is stored and under which jurisdiction’s law it sits. AI sovereignty extends this to inferencing location, encryption key control, model ownership, and the underlying infrastructure’s energy and physical security.
What counts as automated decision making under the new framework?
It covers systems that make or materially influence decisions affecting individuals, such as credit assessments, employment screening, or insurance eligibility, without meaningful human review.
Do small and mid-sized businesses need to act now?
Yes. Building governance before legislation lands is significantly cheaper than remediation after an audit finding or an incident.
Talk to Us Before the Legislation Forces the Issue
Businesses that build governance in now can demonstrate compliance the moment legislation lands. Waiting means running an expensive remediation project against a hard deadline instead.
Acting early also lowers your risk today. Well-governed systems are far less likely to join the 95% of AI projects that never deliver measurable results.
So, ask yourself honestly: could your organisation stand behind a duty of care obligation right now? If the answer is uncertain, our team can assess your current posture and help you build a compliant foundation before the legislation forces the issue.
Get in touch to talk through what this change means for your business.
Sources
1Department of Industry, Science and Resources, “Voluntary AI Safety Standard.”
2Minister for Industry, Science and Resources, “AI consumer safety priorities,” August 2026.
3National AI Centre, “Guidance for AI adoption: implementation guidance | National AI Centre”
4ABC News, “William Yeates charged over deepfake image creation,” 25 February 2026.
5AWDR, “Fake AI case law sinks Fair Work unfair dismissal claim.”
6ABC News, “Deepfake images trial: Antonio Rotondo,” 25 August 2026.
6Prime Minister of Australia, “AI: Australia’s interests,” 15 July 2026.
8National AI Centre, “Guidance for AI adoption: foundations.”
9National AI Centre, “Guidance for AI adoption: implementation.”
10Australian Signals Directorate’s Cyber Security Centre (ACSC), “Engaging with artificial intelligence,” citing its guidance that organisations confirm AI systems meet data residency and sovereignty obligations.
Dan Morgan
Contact us
Talk to an expert.
Let’s talk about where you are and what better looks like for your business.
- 1800 004 943
- Level 15, 2 Market Street, Sydney, NSW, 2000
- Follow us on LinkedIn!







