Intellectual property is safe: The robots will make sure of it.
A federal audit cleared IP Australia's AI patent tools — but the governance gaps it found raise a harder question about the decisions already made.
IP Australia has been using artificial intelligence to help process patent applications since 2018. A new audit from the Australian National Audit Office found this has worked "largely effectively." That finding, however, comes with two recommendations and a set of qualifications that deserve more attention than the headline.
"Assists" and "influences" are not the same thing
The audit, tabled on 29 June 2026, examined four AI tools IP Australia has deployed to assist patent examiners. The tools are designed to surface relevant prior art and information to support human decision-making, not replace it. That distinction matters and the ANAO is careful to preserve it. But the distinction between "assists" and "influences" is thinner than it might appear, and the audit acknowledges that strategic oversight of the AI implementation was "not yet fully established."
Consider what patent rights actually are. A patent is a government-granted monopoly. It gives the holder the exclusive right to make, use, and sell an invention for up to 20 years. Every other commercial actor, every competitor, every follow-on innovator, has to navigate around that right or pay for access to it. The economic consequences of a wrongly granted or wrongly refused patent are not abstract. They ripple through investment decisions, research directions, and market structures. Getting this right is not a bureaucratic nicety — it is one of the core functions of a property rights system.
Those patents are already granted.
The governance gaps matter more than the headline finding
That context makes the governance gaps in this audit more significant than a surface reading suggests. The ANAO's two recommendations target cyber security risk management and strategic oversight. On the first: AI tools processing patent data represent an attractive target. Patent applications contain commercially sensitive technical information filed before any protection is granted. Gaps in cyber security governance are not a technical footnote. On the second: if the organisation deploying AI cannot demonstrate it has a complete picture of how those tools are performing, what benefits they are delivering, and where the risks lie, then "human in the loop" is a design principle rather than a verified reality.
This is where the audit's moderate language creates a slight optical illusion. "Largely effective" and "governance has matured over time" are reassuring phrases. They are also, on reflection, a description of an organisation that learned as it went. IP Australia deployed its first AI tool in 2018. The government's Policy for the responsible use of AI in government did not take effect until September 2024. The APS AI Plan was released in November 2025. For six years, IP Australia was building and refining AI governance in a formal policy vacuum. That the ANAO found the arrangements "largely effective" may say as much about the modesty of the benchmarks available as the quality of the arrangements themselves.
The productivity case is real — but proportionality is the issue
None of this is an argument against using AI in the patent process. The productivity case is real. IP Australia processed nearly 30,000 patent applications in 2024-25, with a median grant time of 2.8 years. Examiners are working through technically complex material at scale, and tools that surface prior art faster or flag relevant classifications are genuinely useful. The question is not whether AI belongs in this process. It is whether the governance around it was proportionate to the stakes.
The honest answer from this audit is: not quite, not yet. The recommendations have been agreed to. The gaps identified are correctable. But the pattern here — deploy first, formalise governance later — is one that the broader public sector is going to have to examine carefully as AI tools move from productivity experiments into the architecture of consequential decisions. Patent grants are a clean, concrete example of that risk. They are legally binding. They shape markets. They cannot simply be walked back because the model that helped evaluate them turned out to be operating without adequate oversight.
The ANAO has done what it should: documented the situation accurately, identified the gaps, made practical recommendations. What the audit cannot do is answer the harder question about the years before the governance matured. Those patents are already granted.
Sources
Australian National Audit Office — Artificial Intelligence Use in IP Australia
Frequently Asked Questions
What AI tools does IP Australia use to process patents?
IP Australia has deployed four AI tools since 2018 to assist patent examiners. They are designed to surface relevant prior art and support human decision-making rather than replace it, though the ANAO audit notes that strategic oversight of these tools was not yet fully established.
Why does it matter if AI governance at IP Australia had gaps?
A patent is a government-granted monopoly lasting up to 20 years — it shapes investment decisions, research directions, and market structures for every competitor who must navigate around it. If the AI tools influencing those decisions operated without adequate oversight, the consequences are legally binding and cannot simply be reversed.
What did the ANAO audit actually recommend?
The audit made two recommendations: one targeting cyber security risk management and one targeting strategic oversight of the AI implementation. Both recommendations were agreed to by IP Australia.
Was IP Australia breaking any rules by using AI before the government's AI policy existed?
Not formally — the government's Policy for the responsible use of AI in government did not take effect until September 2024, and the APS AI Plan was not released until November 2025. IP Australia was deploying and refining AI tools in the absence of a formal policy framework for six years, which means the ANAO had limited benchmarks against which to assess those earlier arrangements.
Does 'human in the loop' actually protect against AI errors in patent decisions?
Only if the organisation can verify that human oversight is genuinely effective, not just a design principle. The ANAO audit found that IP Australia could not yet fully demonstrate it had a complete picture of how its AI tools were performing — which means 'human in the loop' was an intention rather than a confirmed safeguard during the period under review.