The Australian Government has released new guidance specifically addressing artificial intelligence (AI) activities under the R&D Tax Incentive (R&DTI). This update reflects the surge in AI-related claims and provides some clarification on how existing legislative principles apply to modern AI development.
For R&D tax professionals and claimants alike, this is a significant development, not because the law has changed, but because administrative expectations are becoming sharper and more explicit.
Table of Contents
Why this guidance matters
AI is now embedded across nearly every sector from SaaS and fintech to agriculture, healthcare, and advanced manufacturing. The Government has acknowledged this shift, noting that AI is an “emerging and evolving area” within the broader software development landscape.
However, with increased adoption comes increased scrutiny. The updated guidance aims to:
- Help businesses self‑assess eligibility more accurately
- Highlight common risk areas in AI claims
- Reinforce how the existing legislative tests in Division 355 apply to AI
- Importantly, this is not a policy expansion. It is a compliance and interpretation clarification.
AI activities: What’s in scope?
The guidance confirms that AI-related R&D can span the full lifecycle of development. Examples of potentially relevant activities include:
- Designing AI-enabled solutions
- Developing and training AI models
- Testing and evaluating model performance
- Integrating AI into software systems
- Deploying and maintaining AI models in production
- This aligns AI squarely within the software development sector guide, meaning the same eligibility framework applies.
But here’s the critical piece, these activities are not inherently R&D. Eligibility depends on how they are conducted—not what they are.
The core message: AI does not change the law
The guidance strongly reinforces a key principle:
Using AI does not make an activity eligible for the R&D Tax Incentive.
To qualify as a core R&D activity, AI work must still meet the legislative definition under section 355‑25 of the Income Tax Assessment Act 1997, including:
- The presence of a technical unknown
- A systematic progression of work
- An experimental process (hypothesis → testing → evaluation)
- The generation of new knowledge
This principle has always applied, but the guidance makes clear that many AI projects fail precisely at this point.
Common risk areas highlighted
The most practical value of the new guidance is its identification of “red flag” activities that are often incorrectly claimed.
1. Using existing tools and techniques
Simply applying known AI models, libraries, or frameworks (even if new to your team) is not sufficient.
If a competent professional could achieve the outcome using existing knowledge, it is unlikely to be eligible R&D.
2. Routine data preparation
Activities such as:
- Data cleaning and formatting
- Aligning datasets to model input requirements
are typically non‑experimental, especially where the approach is already known.
3. Standard testing and validation
Testing activities that confirm expected outcomes, such as regression or acceptance testing, are generally not R&D.
4. Operational or production activities
Tasks like:
- Monitoring model performance
- Maintaining deployed systems
- Implementing dashboards or alerts
are usually business-as-usual operations, not experimental R&D.
Where AI can qualify
Despite the cautionary tone, the guidance does acknowledge that AI projects can qualify, but only where genuine technical uncertainty exists.
Examples of potentially eligible scenarios include:
- Experimenting with novel model architectures to overcome performance limitations
- Developing new training methodologies where outcomes are uncertain
- Resolving non-obvious technical constraints (e.g. bias, latency, accuracy trade-offs)
The key test remains:
Could the outcome have been known or determined in advance?
If yes → not R&D
If no → potentially eligible (with supporting evidence)
Documentation expectations are rising
Although not explicit in legislative terms, the guidance reinforces a broader compliance trend that evidence is everything.
To support AI claims, businesses should be able to demonstrate:
- The state of existing knowledge (e.g. literature review, benchmarking)
- Clearly defined technical uncertainties
- A documented hypothesis and experimental plan
- Iterative testing, observation, and evaluation
This aligns with broader ATO and AusIndustry expectations that companies maintain contemporaneous records and clearly distinguish R&D from routine development.
Practical implications for claimants
1. Reassess AI-heavy claims
Many claims that focus on “model building” or “AI integration” may need to be reframed around technical uncertainty and experimentation.
2. Tighten activity descriptions
Avoid describing work in terms of:
- Commercial outcomes
- Product features
- General innovation
Instead, focus on:
- The specific technical problem
- Why it could not be solved using existing knowledge
- The experimental approach taken
3. Separate R&D from BAU
AI projects often blend experimentation with deployment. Clear segmentation between R&D and non-R&D activities is critical.
4. Expect increased scrutiny
Given the rise in AI claims, regulators are likely to target this area for compliance activity.
Final thoughts
This new AI guidance is a timely and necessary update in a rapidly changing technology landscape. While it does not alter the law, it sends a clear signal:
The bar for demonstrating genuine R&D in AI projects remains high—and is now more explicitly defined.
For R&D tax practitioners, the message is clear:
Success in AI claims will depend less on the sophistication of the technology and more on the rigour of the experimental process and documentation.
We also note that the new guidance does not mention the Core R&D Activity exclusion around Social Sciences, Arts and Humanities, or provide any commentary regarding situations where companies may be incorrectly classifying AI/ML activities as being related to Information and Computing Sciences (i.e. software claims), instead of framing them around the specific domain of the underlying research.
Navigating grant and incentive applications can be complex, but you don’t have to do it alone. Fullstack Advisory specialises in helping innovative businesses secure funding and scale impact. Book a free consultation today with our R&D Tax Consultants to get expert guidance on your grants and incentives applications to maximise your chances of success.
