More of our clients now arrive with an AI-generated answer to their permit question. That is a good thing: they are informed and asking better questions. But in construction compliance, being almost right is not good enough. A wrong answer can mean unpermitted work, a failed inspection, a rebuild or a dispute. Here is why AI answers need checking, with examples from the past year in Victoria.
Why do AI tools get compliance wrong?
- They learn from a snapshot. A model trained before a rule changed can keep giving the old answer confidently.
- They blend jurisdictions. Building rules vary between Victoria, New South Wales and other states. An answer can merge them without saying so.
- They lean on secondary sources. Much of what is online about building rules is blog content, some of it out of date. Some primary sources, such as official legislation and code websites, are not easily readable by automated tools, so an AI answer may rest on a summary of a summary.
- They produce fluent text, not verified facts. Large language models predict likely wording. When they do not know, they can fill the gap with something plausible.
What does the research say?
The evidence from law, a field close to building compliance, is sobering. A 2024 study by Stanford researchers found that general-purpose chatbots hallucinated between 69% and 88% of the time on specific questions about US federal court cases. A separate 2024 Stanford study of purpose-built legal research AI tools found they still produced incorrect or misleading information between 17% and 33% of the time.
The consequences are real in Australia. In August 2025, a senior barrister apologised to the Supreme Court of Victoria after submissions in a murder case included fabricated quotes and non-existent case citations generated by AI. In September 2025, the Victorian Legal Services Board and Commissioner announced it had varied a solicitor's practising certificate after he submitted AI-generated citations to a court that were inaccurate. If trained professionals can be caught out, so can anyone relying on AI for building rules.
The Office of the Victorian Information Commissioner puts it plainly in its guidance on generative AI: outputs will not always give the correct information, because the tools make predictions based on the prompt and their training.
Five Victorian changes an AI tool might miss
These are real changes from the past two years. An AI trained on older material could get every one of them wrong:
- The regulator changed. The Building and Plumbing Commission began operating on 1 July 2025, bringing together the functions of the Victorian Building Authority, Domestic Building Dispute Resolution Victoria and the domestic building insurance arm of the VMIA. Many websites and tools still refer to the old bodies.
- The building code changed. NCC 2025 has applied to new building work in Victoria since 1 May 2026, with transitional arrangements. An answer based on NCC 2022 may be out of date.
- Home warranty changed. From 1 July 2026, Home Warranty provided by the Building and Plumbing Commission applies to eligible domestic building work over $20,000, replacing the previous domestic building insurance arrangements for new contracts.
- Liquor planning permits changed. Since 1 July 2025, a planning permit is no longer needed just to sell and consume liquor, after Clause 52.27 was deleted.
- Owner-builder thresholds. The current threshold for an owner-builder certificate of consent is work valued at more than $20,000, on a regulator page updated in June 2026. Always check the figure on the official page rather than an older article.
How to use AI safely on a building project
- Use it to learn the concepts and the questions to ask, not to get the final answer.
- Ask it for the source, then open the source yourself. If it cannot point to legislation, a regulator page or the code, treat the answer as unverified.
- Check the date. Victorian building rules often change on 1 July and at other times. Look for the "last updated" date on official pages.
- Confirm the jurisdiction. Make sure the answer is about Victoria.
- Keep project documents out of free consumer tools unless you are comfortable with how the data is handled.
- Let the right person decide. The relevant building surveyor decides building permit questions. Council decides planning questions. A lawyer advises on contracts.
What does responsible AI use look like?
Australian Government guidance on responsible AI adoption emphasises human oversight and accountability: someone must be responsible for what an AI system produces and how it is used. For a building project, that means the practitioner who relies on an AI-assisted output owns the result, just as they would for any other tool.
How AxisPro approaches it
We use AI to work faster, but compliance answers come from primary sources and qualified people. Before we give a client an answer on permits, contracts or regulations, we check it against the legislation and the regulator's own guidance, note the date, and confirm with the building surveyor where the decision is theirs.