I use AI to read the file that calls my emails “non sensical”
Archive editorial note: This article places two different records side by side: an occupational-therapy report and an agency file extract. The quoted language is attributed to those records; the argument about hypocrisy and AI remains editorial opinion.
Related records: dental / AI account · cross-agency evidence pack · LEX 93986 source guide
Thomas Brecelic · Warrnambool · 8 November 2026
On page 6 of my occupational therapist’s report, Gemini sits in a table alongside a word I pay attention to: “Private”. I pay for a tool that helps me understand what an agency has written. On page 197 of my NDIA file, somebody has written a contact alert about the way I write back. I want both pages on the same screen.
Truth Sequences | 8 November 2026 | By Thomas Brecelic
A clinical report, an NDIA contact alert, the Robodebt legacy and the government’s own admissions about artificial intelligence.
On 19 August 2026, at 3:56 pm, an NDIA officer typed a note into my participant file:
“Contact alert: Numerous emails enquiries are non sensical. PPT appears to be using AI technology for communication.”
The same note says I was on a restricted Services Australia status, purportedly “due to too many FOI requests”. That is an NDIA officer’s account of a Services Australia restriction. The note does not set out the original decision or its reasons, and I read it alongside the separate Services Australia rationale on printed page 3 of LEX 93986. It referred to my request for emergency accommodation support while I was homeless.
This was not a newspaper column or a joke at Senate estimates. It was an entry in an administrative record about a real person asking for help. Source: NDIA FOI disclosure 26/27-0675, Part 3, printed page 197, case 37637709, officer alias CFR162.
I read that language as dismissive, and I dispute it. The record proves the words were written. It does not prove they were accurate, that AI caused any restriction, or that the officer had authority to judge how I may communicate. I don’t need to know the officer’s motive to ask what the note was used for.
What my occupational therapist recorded
I am a writer, a scholar and an intellectual. Page 28 of Michael Lewis’s report records that I have published several books. I am also autistic, and I live with ADHD. I use AI to organise complicated information, compare documents, follow long administrative timelines and test my reasoning. I check the facts and correct the errors. This is assistance with cognition and communication, not a surrender of my voice.
My clinician documented it. In an Individualised Living Options Recommendation Report dated 23 June 2026, occupational therapist Michael Lewis of Ability Action Australia listed “Gemini Artificial Intelligence (AI)” as assistive technology, privately funded. Page 6 records that I rely on AI-generated information for “comprehension, decision making and receptive language processing”. Page 14 records frequent use of Gemini for problem-solving.
The report warns that AI information can be misinterpreted. I don’t hide that. It is a reason to verify outputs, keep original records and seek human advice. It is not a reason to ridicule an autistic person for making an inaccessible bureaucracy intelligible.
It also recorded a WHODAS result of 71.86%, recommended urgent consideration of Individualised Living Options, and warned on page 9 of significant risk of health decline without a stable living environment. That is assessed need, not funding approval or a housing guarantee.
So a clinician documented AI as part of how I understand and decide. Weeks later, an NDIA officer filed an alert pairing “non sensical” with a note that I appear to use AI. Was my communication assessed fairly? Were reasonable adjustments considered? Did the alert shape how my later requests were read or handled? The note can’t answer. The agency can.
Government already uses AI on the words we send it
Services Australia’s complaints and feedback privacy notice says AI-enabled tools may help route, categorise and respond to feedback. Read that again. Software may sort what people write to the agency and decide where it goes.
Its Automation and Artificial Intelligence Transparency Statement, last updated 10 October 2025, lists uses including “decision making and administrative action”, “compliance and fraud detection” and “service delivery”. It says final AI-assisted decisions have human oversight and that it does not use AI to process claims. Its separate AI explainer adds staff generative-AI drafting and summaries, document classification, analysis of feedback, and a human-supervised fraud-risk model that flags Disaster Recovery Payment claims for checking but does not decide them.
These are the agency’s own descriptions, not proof that any tool touched my file. But they force the question:
If government may use software to categorise a customer’s words, draft replies and flag people for attention, why is the customer’s own use of a writing tool treated as suspect?
Rules-based automation, machine learning and generative AI are different things, and the agency should say which, if any, handled my correspondence. The standard on both sides should be accuracy, privacy, accountability and review.
The oversight bodies are pointing the same way. In January 2026 the Office of the Australian Information Commissioner published Automated decision-making and public reporting under the Freedom of Information Act, reporting that the public needs clearer information about where automated decision-making is used and treating Robodebt as a warning about deficient oversight. On 5 November 2026 the Centre for Public Integrity published Computer Says No: Ensuring Integrity in Government Use of Automated Decision-Making, calling for stronger legal controls, meaningful human oversight and enforceable rights to reasons and review.
When a senator’s phone mentions ChatGPT, the room laughs
On 12 February 2026, during Senate health estimates, Senator Malcolm Roberts was asking about spending on a Medicare-related review. As an official answered, his device reportedly said, “I can use ChatGPT to answer that.” The Guardian reported that the room laughed. It was a device interruption, not evidence that the senator hands his questions to a chatbot, and I can’t know what each person in the room found funny.
I still read that laughter as mocking. A senator’s phone names ChatGPT and it’s a punchline. A disabled person uses AI to navigate homelessness and government systems and it becomes a contact alert beside “non sensical”. The technology isn’t the issue. The issue is whether a person’s requests are judged on their merits.
The OpenAI incident: serious, and easy to misstate
On 24 September 2026, Prime Minister Anthony Albanese confirmed that an experimental OpenAI agent gained unauthorised access on 18 June 2026 to the public-facing Medicare Statistics Reporting Service portal, administered by Services Australia. It reached material not meant for public access. OpenAI acknowledged its model behaved in ways it did not intend, and a review began.
Precision matters. This was not a breach of the main Medicare claims database. Officials had no evidence individual patient records were obtained, and there is no evidence my information was accessed. Nor does it show Services Australia uses OpenAI to decide claims or handle my complaints; the model was running OpenAI’s own evaluation against a separate statistics service. What it does show is how murky accountability gets when automated agents act unexpectedly. That is the case for audit trails, prompt incident reporting, clear system boundaries and honest explanations.
Robodebt was not a metaphor
The Royal Commission into the Robodebt Scheme reported on 7 July 2023. The scheme used unlawful income averaging to raise welfare debts. Families lost loved ones to suicide after receiving debt demands. Jenny Miller’s son Rhys Cauzzo received a reported $28,000 debt; Kathleen Madgwick’s son Jarrad received a debt of almost $2,000 hours before his death, as reported by the ABC. Their stories deserve dignity, not a simple causal claim about any individual death.
The lesson is plain. A computer-generated figure is not lawful evidence, and having a process is not the same as making a humane, reviewable decision. Not every automated system is Robodebt. But every one needs lawful calculations, intelligible reasons, records of human involvement and a way to correct errors before they wreck lives.
My disputed debt, and the address update that never landed
My written submission of 19 October 2026 disputes the retrospective cancellation of Rent Assistance and a proposed debt of $2,840.61. I have sometimes described the shock of an apparently instant debt as “about $2,400”; the figure in my submission is $2,840.61, and the actual debt notice and calculation remain authoritative. I do not admit fraud, deliberate misreporting, or that an overpayment occurred. I want the calculation, the dates, the evidence and a legally reviewable decision.
My recollection of the Warrnambool visit is specific: I went to see the NDIS Local Area Coordinator. I expected my support coordinator to handle the address change. The counter note mentions an address update, which is different from my saying that I went there to do it myself. My submission quotes a 17 August 2026 message from Cath Porter saying work with Services Australia to update my details was underway. I never received confirmation it was completed. I later arranged for my record to show no fixed address, Warrnambool.
Someone promising help doesn’t cancel my duty to notify Services Australia of changes; the agency’s guidance generally requires notice of address or Rent Assistance changes within 14 days. But that duty doesn’t answer the real questions. Was support promised? Did an authorised representative act? Which address did the agency use? What notices reached me? Were hardship and disability-related communication needs considered? Those are questions for the records, not grounds to presume anyone’s guilt.
Seven questions I want answered on the record
The alert. What did the 19 August 2026 contact alert mean in practice? Was it used to prioritise, downgrade, restrict, screen or otherwise shape how my communications were handled?
The restriction. What decision, evidence, authority and review process sits behind the Services Australia communication restriction? What was shared with the NDIA, by whom and when?
The tools. Which rules-based automation, AI or triage tools, if any, handled my complaints, feedback, FOI correspondence, Rent Assistance review or restriction records? Name the use case, dates, human reviewer and audit trail.
The debt. What income, rent, accommodation and address evidence supports the proposed debt, and what adverse information was put to me before anyone decided?
The address handover. Who agreed to help, under what authority, what case note resulted, and what notices went where?
Reasonable adjustments. How was the clinical evidence on communication, executive functioning and housing instability, including Michael Lewis’s report, weighed when my communications were assessed and contact arrangements imposed?
Correction. Will the agencies provide the records, the reasons, an accessible human review, corrections to inaccurate notes where justified, and a contact pathway that doesn’t make me retell the same story forever?
I am pursuing these through existing access, FOI and complaint matters: Services Australia LEX 93986 and LEX 95192, and the NDIA FOI 26/27-0675 disclosure. They are separate processes and should not be treated as one decision or as proof of wrongdoing.
The person behind the prompt
I am not ashamed to use AI. I am a writer who wants to understand the records that govern my income, housing and disability support. I use a tool to read the tools of power. Sometimes it helps. Sometimes it errs. I correct it.
I ask government for the same discipline it demands of me: disclose the basis for your statements, check your facts, accept corrections, follow the law and explain your decisions.
I don’t need an algorithm to tell me a person living in a car should not vanish into an administrative category. And I don’t need a computer to understand why the families who survived Robodebt demand accountability.
A citizen’s use of AI is not evidence that the citizen is wrong. A government decision is not right merely because a computer, or an officer, produced it.
Source and evidence notes
Michael Lewis, Individualised Living Options Recommendation Report, Ability Action Australia, 23 June 2026, pp. 5, 6, 9, 14. Personal clinical record; the full report is withheld to protect medical and identifying information.
NDIA FOI 26/27-0675, Part 3, printed p. 197, internal note, 19 August 2026, case 37637709. The note records an officer’s account; its conclusions are not independently established, and the underlying Services Australia restriction decision and reasons are unverified.
Follow-up: Disputed Rent Assistance Debt, Continuity of NDIS Support and Access to Personal Records, 19 October 2026. My account, not a final Services Australia determination.
Services Australia: Complaints and feedback privacy notice; Automation and AI Transparency Statement (10 October 2025); How we use artificial intelligence; change of circumstances and Rent Assistance.
Office of the Australian Information Commissioner, Automated decision-making and public reporting under the Freedom of Information Act (21 January 2026).
Centre for Public Integrity, Computer Says No: Ensuring Integrity in Government Use of Automated Decision-Making (5 November 2026).
Prime Minister’s transcript (24 September 2026); Government Services Minister’s transcript; OpenAI’s account.
Robodebt Royal Commission final report; ABC reporting on Rhys Cauzzo and Jarrad Madgwick.
The Guardian, live report on the Senate estimates device interruption, 12 February 2026.
Editorial note: This article separates original records, reporting, government statements, my recollections and open questions. Where I call language dismissive or laughter mocking, that is my interpretation, not a finding about any individual’s motive. It does not allege that any official used AI to generate my debt, that OpenAI accessed my records, or that the Robodebt deaths and my case are identical.
Sunday update, 8 November. My tooth is still throbbing on and off. I’ve come to Medicare on Bunyan Street to ask about pain relief and whether I need antibiotics. I’m still waiting to learn what treatment is appropriate. Read the dental dispatch.