The promise of generative artificial intelligence (AI) was a new era of efficiency.

In some respects, it has done that. Decades of meeting minutes can be fed into an AI program with a request to find any approvals that have been given for a particular lot to make improvements to the common property, or to identify and summarise decisions made about a roofing membrane.

But it has been used as a substitute for human reasoning, analysis and drafting to prepare correspondence and submissions to argue with other stakeholders in a strata community.

A problem has emerged from this: strata managers, committees, lawyers and adjudicators are now dealing with a proliferation of AI-generated complaints and submissions riddled with inaccuracies, arguments stretched beyond what is reasonable, and “hallucinations”.

In a recently published decision, an adjudicator dismissed an adjudication application and made a $2,000 costs order against the applicant because the way they conducted their case amounted to an abuse of process that had wasted the resources of the Commissioner’s Office and put the body corporate to considerable defence expenses.

The adjudicator discovered that the applicant used AI to generate their submissions.

The limits of AI-generated content

AI has made its way into functions and tools used each day. The world’s largest internet search engine – Google – provides an AI Overview to any query put to it.

We “googled” this query: “does an AI program like ChatGPT reason and analyse data, or simply identify patterns”? The AI Overview was the top result returned and gave this summary:

“AI like ChatGPT primarily identifies patterns, but it can simulate reasoning and data analysis by using these patterns to make predictions and generate responses. While it can summarize data, find trends, and even write code to perform analysis, it does not “reason” or “understand” in a human sense; its “analysis” is a sophisticated form of pattern matching learned from its training data.”

We then went straight to the source and raised the same query with ChatGPT.  Its answer included this explanation:

“While they do perform some form of reasoning in a probabilistic sense—connecting pieces of information and making inferences—they do not reason or analyse data in the same way humans do with conscious understanding or logical deduction.”

Those who fear that AI will replace jobs held by people in professional fields (such as lawyers) might take comfort in reading that an AI program does not reason and analyse information with the conscious understanding of the information held by some humans. A computer program cannot reason and empathise like a human can.

The legal advice lawyers give may not be what their client wants to hear. That’s a part of the legal profession. A lawyer’s first duty is to the court – not their client. It is our duty to assist in the administration of justice by helping the court, tribunal or adjudicator make the correct decision, rather than blindly advocating for a client’s cause in the hope they get the outcome they want.

An AI program is not under that duty. It will generate an argument it is asked for.

We asked ChatGPT to develop an argument about a body corporate of a community titles scheme in Queensland having no duty to maintain common property in good condition. At a surface-level, the argument it generated looked professional, well-constructed and well-reasoned. This is our summary of that argument:

• The BCCM Act sets out general objectives for bodies corporate to administer and maintain common property but does not expressly impose an unconditional or absolute duty. The duty to administer common property is broad and flexible, giving a body corporate a discretion in prioritising resources and determining what level of upkeep is reasonable or proportionate.
• The body corporate acts as an agent of the lot owners so its capacity to act (and therefore fulfil a duty to maintain) is limited by the resources and resolutions approved by those owners. Imposing a positive duty to maintain common property would contradict the collective, democratic nature of the scheme’s governance, so the duty is only conditional, not absolute.
• The BCCM regulation modules do not provide an objective criteria for what maintaining common property “in good condition” means. Accordingly, what is “good condition” is subjective, so the duty cannot be enforceable.
• There are good public policy reasons to not interpret this duty too strictly – it would make a body corporate strictly liable for natural deterioration, misuse by owners or funding constraints.

We can confidently say none of those arguments are legally correct. But upon reading those arguments, those without our experience and expertise may hesitate and second-guess their understanding of this primary duty held by a body corporate.

We wanted to see how far the AI-program would go with this argument. We followed it up with this query:

“The situation I have is that there is a common property wall with faded paint. A lot owner says it is in disrepair and needs to be repainted to comply with the duty to maintain common property in good condition. How should I argue against that duty applying?”

ChatGPT gave an argument which it self-described as “clear” and “persuasive”. It explained that its argument and interpretation of the BCCM Act and Standard Module is supported by:

• The Rocks Resort [2009] QBCCMCmr 262, repainting for aesthetic uniformity was classified as an improvement, not maintenance, where the existing paint still adequately protected the building.
• Sunshine Beach Apartments [2015] QBCCMCmr 56, it was found that normal fading of paint did not constitute disrepair requiring immediate maintenance.

We clarified that the situation is not just about mere cosmetics because the paint is bubbling, peeled in other parts and generally looking awful, and asked if this changes any of those arguments.

ChatGPT conceded that the protective function of the paint may be comprised so a duty to maintain likely arises but assured us that we could still argue against the repainting duty if we can show that the damage is minor or localised, the wall remains weatherproof and structurally sound, and the issue is primarily visual rather than functional. It said this reasoning is supported by these adjudication decisions:

• Ocean Isles [2013] QBCCMCmr 198, it was stated that “repainting becomes maintenance only when the existing coating no longer provides protection to the building fabric.”
• Couran Cove [2012] QBCCMCmr 394 found that “peeling or flaking paint exposing the substrate may amount to a failure to maintain,” but aesthetic deterioration alone does not.

We wondered whether the AI-generated response was misapplying (or ‘stretching’) those adjudication decisions. We decided to read all four of those cases.

The problem we encountered is that none of those cases exist.

The Rocks Resort and Couran Cove are well-known strata communities on the Gold Coast. Both have been frequent features in litigation before the Commissioner’s Office. Ocean Isles is lesser-known, but The Ocean Isles has one adjudication decision to its name. However, the cited decisions for these schemes don’t exist.

Sunshine Beach Apartments sounds like a good name to give a strata scheme, but online search records (austlii.edu.au) reveal no adjudication decision exists for that scheme (if the scheme even exists).

This isn’t a simple misunderstanding of the law. Those cases were “hallucinations” presented with a veneer of plausibility.

The prevalence of AI-generated material

It has become a regular occurrence in our legal practice to encounter AI-generated content. It isn’t confined to submissions in adjudications or other litigation. It is more regularly occurring in e-mails and letters, such as a caretaker’s response to committee queries or a complaint from a lot owner directed to a strata manager / committee.

It is relatively easy to identify it as AI-generated content:

• the formatting used replicates the formatting produced by AI services;
• the content misquotes legislative provisions or cases;
• untenable or irrelevant arguments are made;
• key detail is missing from factual matters or is simply misdescribed;
• the drafting style or language used is inconsistent with previous written communications the person has sent; and
• the letter has an artificial (and sometimes insincere) ‘feel’ to it.

The consequences of AI-generated content

The use of AI-generated content is intended to make the process of raising (or responding to) a query or complaint quicker and more efficient. Ironically, this generates an inefficiency, particularly in legal disputes. Someone will need to:

• manually search for every case and legislative reference, one by one, to see if they are real; then
• identify which parts of the argument are based on real law (and are just wrong), and which are stretched beyond reasonable limits or are based on fantasy.

That ‘someone’ should be the person who wants to use the AI-generated content, if they were exercising skill, care and diligence.

But often that ‘someone’ is the person on the receiving end.

In Sky Gardens [2025] QBCCMCmr 373, it was the adjudicator and the body corporate having to deal with an adjudication application, supporting submissions, and ‘information’ produced by a lot owner who had used AI to generate their material. The adjudicator had this to say at [68]-[69] of their reasons:

“The way the applicant pursued this application, including the continual reliance on non-existent or irrelevant case law and legislative provisions, and incorrect or unsubstantiated claims about the submitted evidence, has impeded the determination of this application. It has certainly required more time for me to investigate the application, and it undoubtedly increased the costs for the body corporate to respond to the application.

I am satisfied the application is misconceived and without substance. Moreover, I consider the scale of the deficiencies in the submitted material amounts to an abuse of process. The body corporate has been put to considerable expense in responding to this application. The resources of this office have also been wasted. I am satisfied an award of costs is warranted.”

Criticisms of this kind are being given across all judicial forums. For example, in LJY v Occupational Therapy Board of Australia [2025] QCAT 96, Judge Dann made these comments at [26]:

“It is important that Ms LJY, and other litigants before the Tribunal, understand that including non-existent information in submissions or other material filed in the Tribunal weakens their arguments. It raises issues about whether their submission can be considered as accurate and reliable. It may cause the Tribunal to be less trusting of other submissions which they make. It wastes the time for Tribunal members in checking and addressing these hallucinations. It causes a significant waste of public resources.”

Concluding remarks

Those who use AI-generated content when corresponding with other stakeholders in a strata community, or making submissions to the Commissioner’s Office, should appreciate that they will quickly lose credibility and the trust and confidence of others if they don’t use it carefully and diligently.

It might seem that in the heat of a dispute or disagreement, there is little trust or confidence to be lost given the already frayed relationships. But living and working together in a strata community goes beyond an exchange on one issue in dispute. More issues may arise in the future. Everyone wants their queries and concerns to be taken seriously, but that requires trust and confidence that they ought to be taken seriously.

Article Contributed by Jason Carlson, Partner at Chambers Russell Lawyers.

The post THE USE OF AI-GENERATED CONTENT IN STRATA DISPUTES appeared first on Smart Strata | Body Corporate Management.