New AI Guidance Released for the R&D Tax Incentive
Author: Shaun van Dijk
The Department of Industry, Science & Resources (DISR) has released new guidance on Artificial Intelligence (AI) related R&D activities under the Research and Development Tax Incentive (RDTI).
The guidance is intended to help businesses understand how existing RDTI eligibility requirements apply to AI-related projects, including machine learning, generative AI, computer vision and predictive analytics activities.
AI Is Not a Separate Category of R&D
Importantly, the guidance does not introduce new rules, eligibility criteria or legislative requirements. The existing R&D Tax Incentive legislative requirements are the primary reference for eligibility and apply to AI-related activities in the same manner as any other industry or technology area.
The guidance represents DISR's interpretation of how the existing law may apply to AI-related activities. It also acknowledges that the information provided is general in nature.
What AI Activities May Be Eligible?
The new DISR guidance confirms that AI-related activities may be eligible where businesses are seeking to resolve genuine technical challenges through experimentation, such as:
Developing or improving machine learning algorithms or models.
Testing alternative approaches to achieve a technical outcome.
Addressing challenges relating to model accuracy, reliability, scalability or performance.
Creating new methods for processing, classifying or interpreting data.
As with any R&D claim, the focus is on establishing (1) the technical uncertainty being addressed, and (2) the need for experimental activities to be undertaken to resolve or overcome this technical uncertainty.
What Activities Are Less Likely to Qualify?
The guidance also highlights the types of AI activities that are generally unlikely to satisfy the R&D criteria, including:
Routine model deployment, monitoring and maintenance activities – e.g. implementing logging, alerts, dashboards and performance monitoring using established tools and practices.
Standard data preparation and processing activities – e.g. data cleaning, formatting, transformation or alignment that follows known approaches and the required outcome is understood in advance.
Verification and quality assurance testing – e.g. regression, acceptance and functional testing designed to confirm a system performs as expected.
Routine model tuning and optimisation, where parameters are adjusted using established techniques.
Integration of AI models into software products or business processes – e.g. connecting known model outputs to applications, workflows, dashboards or decision-support systems.
Implementation activities focused on achieving a known outcome where the activities do not require experimentation to resolve genuine technical uncertainty.
While these activities may be innovative from a business perspective, they will not necessarily satisfy the legislative requirements for R&D.
In short, simply using AI, machine learning models or generative AI tools does not automatically make an activity eligible for the RDTI. Whilst these types of AI-related activities may be complex, if the underlying challenges can be addressed using existing knowledge and solutions, they are unlikely to qualify as eligible core R&D activities under the RDTI.
Why This Matters
The release of AI-specific guidance reflects the growing importance of AI across the Australian economy. However, it also highlights the challenges and increasing complexity of applying a long-standing legislative framework to technologies that are evolving at an unprecedented pace.
Many AI projects sit within a grey area between routine software development and genuine experimental R&D. Determining where that line is drawn, and which activities fall either side, is likely to remain one of the more contentious aspects of the program.
Documentation Remains Critical
As with all R&D activities, businesses should ensure they maintain contemporaneous records demonstrating:
• The existing knowledge at the outset of the R&D and the technical uncertainty being addressed.
• The hypotheses being tested.
• The experiments undertaken.
• The results observed and conclusions reached.
For AI projects in particular, documenting the experimental process will be critical in distinguishing genuine R&D from routine development or implementation activities.
Our View
The release of dedicated AI guidance is a welcome step and, whilst the examples provided are fairly simplistic, they should assist businesses navigating this increasingly complex area of the R&D Tax Incentive.
However, the key principles remain unchanged. AI related activities are not automatically eligible R&D activities under the RDTI, nor will they be automatically excluded or deemed ineligible. As with any other sector or technology area, RDTI eligibility depends on whether a company is undertaking systematic experimentation to resolve technical uncertainty and generate new knowledge, and whether these attributes can be evidenced.
As AI technologies continue to evolve, we expect this area to remain a significant focus for both claimants and regulators, making careful R&D project planning, governance and documentation more important than ever.