DTA Deadline for AI in the APS
Government agencies are facing a tight deadline to incorporate Artificial Intelligence (AI). Under the Digital Transformation Agency’s (DTA) policy for responsible use of AI in government (effective 15 December 2025), non-corporate Commonwealth entities must:
Develop and communicate their strategic position on AI adoption to staff by June 2026
Publish a transparency statement outlining their approach to AI and must review and update this statement annually
Implement mandatory training on responsible AI for all staff by December 2026
create a register of all in-scope AI use cases by December 2026 and share this register with the DTA every six months.
Beyond these administrative requirements, agencies must align their adoption with Australia’s eight AI Ethics Principles, which include values such as Human, Societal and Environmental Wellbeing, Fairness, Transparency and Explainability, and Contestability. These principles require practical implementation through ‘essential practices’ such as maintaining human control through override points and ensuring accountability across the entire AI supply chain. This is critical to avoid programs like Robodebt, which did not have the human interventions built in from the outset.
Should I be worried about this?
The debate around incorporating AI into the public sector tends to polarise into two camps: a race toward adoption, or a stance against job replacement. Framed that way, it is unsurprising that the debate has generated significant fear.
But the best way to think about AI is more nuanced. The ‘gift of time’ AI provides should be spent on strategic thinking and honing human skills, not just on speeding up administrative tasks. By adopting responsible AI use now, agency employees aren't just protecting their roles, they are potentially building a more defensible service. If this is coupled with a solution that allows collaborative workflows to mitigate bottlenecks, then this is a great outcome.
Put simply, as we’ve said in many of our previous articles, AI is an accelerator. It reduces the time spent on routine administrative tasks, so that more time can be spent on high value-added tasks. AI doesn’t replace people; it allows people to better use the skills for which they were hired. Agencies that adopt AI with a collaborative workflow system will thrive over those that don’t.
But what if AI is used only to speed up discrete steps in a digital system without the systems thinking and workflow automation that ensure your Delegate has time to check, consider and make an informed decision? What if the content pushed produces more than any decision-maker has time to consider? What if the expectations to manage the content increase?
So, what if AI doesn’t give me the ‘gift of time’?
Psychosocial hazard regulations are not new; however, the states are starting to focus on WHS risks in the AI and Digital Systems space. The Work Health and Safety (Digital Work Systems) Act 2026 changes how NSW businesses must approach workplace technology. The law makes it clear that AI, algorithms and automated systems can create workplace risks, and they must be managed.
Under this Act, organisations must ensure that digital systems do not expose workers, amongst other things, to:
excessive or unreasonable workloads imposed by algorithms;
performance metrics that create undue pressure; or
discriminatory decision-making by biased AI systems.
If an algorithm creates impossible workloads, the employer is legally responsible regardless of whether the AI software was created by the organisation or licensed from a third party.
A policy and training sessions are not enough: regulators across states are focusing on how work is actually performed. Low job demands (where judgement is eliminated and autonomy is lost) can also create risk to health and safety. Machine-paced work which is not monitored and creates unrealistic work pressures will produce psychosocial safety risks.
How can these risks be removed or mitigated?
Mitigating these risks requires clear visibility and control over workload. A whole system approach, supported by a technology solution that manages workflow, workload and performance is the right approach. Human autonomy to interrogate the system and make changes is essential to ensure outcomes are met and improvements made where needed.
We have been helping clients with producing policies, providing training and auditing AI for use and legal risk. We believe the real opportunity is with designing and implementing workflow solutions specifically for these complexities, ensuring compliance, transparency and autonomy are embedded into the system and all decision-making processes. This lets agencies move from mere document production and content creation to a model where strategic positions actively guide behaviour. Synlaw's workflow solutions are built around how government operates and support enablement of outcomes without bottlenecks that AI solutions alone produce.
Agencies can then adopt the DTA's new requirements in lockstep with the evolving regulatory framework using them not just to protect employees, but as a foundation for more transparent and efficient outcomes.
As Bobbi Campbell said last edition; human judgement and context can’t be replaced by AI alone, despite the fact that many AI tech companies are saying it can. That said, she is seeing that a system could support context, if context is not considered a single ingredient. Tech and talent stacks within a real Reg Tech environment are available now. More on who’s doing something about the ‘context’ in our next edition.