When AI Writes the Report: Lessons From an NDIS Tribunal Decision

In August 2025, the Administrative Review Tribunal upheld the NDIA's refusal of extra supports for an NDIS participant. The Australian Financial Review reported it as a bid to boost a roughly $200,000 package, rejected after AI use was found.

The headline makes it sound like a case against AI. The decision itself is a case about accountability.

What happened

The participant sought a substantial increase in supports, including hundreds of hours of physiotherapy and exercise physiology, plus occupational therapy, psychology and support worker hours. Several allied health practitioners provided reports.

The NDIA raised concerns that some reports had been produced with AI without adequate quality control. Under questioning, two physiotherapists' reports were found to have been drafted with an AI program.

One report justified supports by referring to the participant's rural location. She lived in an urban area, and the author couldn't explain where that came from. It also contained a key phrase its author didn't remember writing; she couldn't rule out that the AI or a supervisor had added it. The tribunal couldn't follow the reasoning behind some conclusions. And the reports rested largely on material the participant had supplied and limited telehealth contact, in one case a single consultation.

The tribunal gave those reports little weight and affirmed the NDIA's decision.

What the tribunal said about AI

It had no objection, in principle, to a practitioner using AI to put their own observations and assessments into writing. But the practitioner must stay responsible and accountable for what the report says. That's the same standard Ahpra sets for every registered practitioner.

Why it matters

The cost fell on the participant. Reports meant to support her case ended up weakening it. That's the real risk of careless AI use: not embarrassment for the clinician, but harm to the person relying on them.

There's a wider cost too. Cases like this invite funders and insurers to scrutinise allied health reports more heavily, which means more paperwork for every clinician, including the careful ones.

Lessons for anyone writing reports

Your name means your report. If you sign it, you must be able to defend every sentence, under questioning, without notes.

AI writes up your assessment. It doesn't do it. The observations and reasoning have to come from you. A report recommending hundreds of hours of therapy needs an assessment that can carry that weight. Thin evidence doesn't get stronger because it's well formatted.

Read it the way a tribunal would. Every fact should trace to your own notes, every conclusion should be explainable in your own words, and every citation should be checked. Watch for the tells: facts you never recorded, phrases you'd never use, generic justifications that could fit any patient. If you find one, check the whole report again.

Keep the trail. Keep your notes and drafts, and know what the AI changed. Where AI use is material, be open about it. If a supervisor edits your report, agree with every change before it goes out under your name.

Give high-stakes reports a second reader. NDIS, insurer, workers' compensation and medicolegal reports deserve more than a skim. A second clinician's review is a small cost against a failed claim.

The bigger picture

I wrote about this case for the APA's InMotion in late 2025, and I keep coming back to it. AI is excellent at turning your thinking into clear, well-structured writing. It's dangerous when it starts doing the thinking. The tribunal drew that line clearly, and so should we.

---

For the profession's own principles, see the APA's statement on AI in physiotherapy.