NDIA Accountability Project

This project is in DRAFT

Submission 3216 — Dr Tamsin Clarke — NDIS Future Generations Bill

‹ PrevPage 1 of 4 · Source p. 1Next ›

1 June 2026

Committee Secretary

Senate Standing Committees on Community A>airs

PO Box 6100

Parliament House

Canberra ACT 2600

By email: community.a>airs.sen@aph.gov.au

Dear Committee Secretary

Submission in relation to National Disability Insurance Scheme Amendment (Securing the NDIS for Future Generations) Bill 2026 (‘the Bill’)

Background

I am writing as the friend of an NDIS participant and as a lawyer concerned with the undermining of proper administrative decision-making through the use of Artificial Intelligence. Some of the material in this submission may therefore also be covered in her separate submission. I am very concerned that Schedule 3 of the Bill contemplates the use of ‘automated decision making within the NDIS.’ I strongly believe that automated decision making should not be used by NDIS. To the extent any automated decision making is used, human review should take place and appeal against such decisions must be possible.

Why AI decision making should not be used by NDIS “We tend to assume artificial intelligence is super smart. But it is artificially intelligent. That word matters. It gathers patterns. It predicts likely sequences. It fills in blanks. It does not understand consequences. It does not weigh nuance. It does not experience doubt. It does not carry the emotional burden of being wrong.

And it does not hesitate unless we design systems that require it to.

So the question is not whether AI is impressive. It is. The question is whether we are ready to delegate authority to systems that are optimized for fluency, not for judgment.

… Used carefully, it can be helpful. But helpful is not the same thing as trustworthy.

… Before we hand over more authority, we should be asking harder questions about accountability. And before we surrender our judgment to systems that predict patterns, we should remember that judgment—the kind that pauses, weighs, and sometimes says “I’m not sure”—is still profoundly human.

2

And that is something we cannot afford to automate.”1

  1. Like my friend, I am concerned:

  • at the concept in the new section 9B definition of functional capacity of excluding the impact of the person’s environmental and personal circumstances; and

  • that under section 9B(3) rules made for the purposes of that subsection may prescribe: … “(b) matters that may, must or must not be taken into account for the purposes of subsection (1); or (c) circumstances in which a matter relevant to the application of subsection (1) is taken to exist or to not exist in relation to a person.”

  1. Similarly, the proposed changes in Part 3 of Schedule 1 are of concern, apparently limiting services to those ‘core’ disabilities which were the historic reason for the participant being eligible for NDIS participation – despite the fact that one disability can lead to another and that we all become increasingly disabled as we age.

  2. Since the NDIS was implemented, local governments and private charities have withdrawn a raft of services that were provided across the board to disabled people, the elderly, and those with temporary illnesses. This has been a disaster for people who were not su>iciently disabled to receive NDIS support. Now the idea seems to be that even an NDIS participant will not be supported through the NDIS in relation to functional problems which are not part of their original ‘core’ disability. Like non-NDIS participants, they will now have vastly reduced options for assistance in relation to those functional problems.

  3. These new provisions seemed aimed at assessing a person’s functional capacity on an entirely theoretical basis, quite apart from their real life situation. That should never happen within NDIS. NDIS has been created to support individuals who have serious functional problems. Each individual is unique and has the right to be treated with dignity. Imposing some kind of theoretical functional assessment on a participant, after disregarding their situation and other relevant matters, was not what the NDIS was set up for, and is a very bad way of responding to the fraud which has been identified. Theoretical functional assessment might be helpful for ergonomics, economics or clinical medicine, but has no place in a system which is meant to respond to an individual’s needs.

  4. It is not by taking away human judgment and the rule of law that fraud can be countered, and participants helped, but by enshrining these features. Like common law, NDIS is not rule by algorithms but rule by sensitivity, experience and precedence. To remove human judgment and replace it with AI is to replace compassion with a model that fits no-one. Policy issues require connection to the real world (as Professor Francis Fukuyama says2).

  5. To quote another commentator3: “AI systems can generate output that looks like judgment. They can weigh factors, consider trade-offs, produce recommendations that sound thoughtful.” But because those systems cannot understand the impact of their

1 “The Resister’s Report” at https://theresistersreport.substack.com/p/ai-isnt-smart-its-just convincing 2 “What AI Hypists Miss” Francis Fukuyama at https://www.persuasion.community/p/ai-cant-deal with-the-real-world 3 Email from Srini Rao 21 April 2026: “You cannot read your way to better judgement”

3

output, and there is no negative consequence for the system in making an incorrect call, what the systems are providing is not real judgment. “This is not a temporary limitation in AI that will be solved with better training. It is definitional.” AI systems can never exercise real - that is human - judgment. “They are structurally excluded from a category of moral knowledge upon which some of our most consequential decisions depend.”4 Speed is useful when the task is mechanical, says the writer, not when the task is moral.

  1. That these concerns are valid is confirmed by reports (Guardian 24 February 2026) of serious concerns relating to incorrect outputs from the Federal Government’s Integrated Assessment Tool to determine eligibility for aged care services and funding levels. Asessors have been given limited scope to overturn incorrect outcomes.5 It is clear that a system which cannot itself understand the real world consequences of its decision-making is not, without extensive human supervision, appropriate to a system such as NDIS.

The right question is not whether AI can help. In many domains, it probably can. The right question is what proof must exist before AI is permitted to influence decisions with serious human consequences.

Any AI system used in targeting support should require auditable data provenance, adversarial testing under realistic conditions, clear and exercisable override authority, strict logging, post-action review, and legal accountability that cannot be transferred to a vendor.

… The ceiling should be a democratic rule: the more severe the consequence, the less acceptable automated judgment becomes. A model can recommend a spare part. A model can summarize documents. A model can flag anomalies. 6

But an AI model should not decide who receives NDIS support, nor what the medical data about a participant means.

Part 4 of Schedule 1- Support determinations

  1. I submit that legal services should be included in the NDIS price list. They are not. Accounting services are covered, physiotherapists, nutritionists etc – however, day in day out, participants need lawyers to advise them on a whole host of contractual issues which have real life legal consequences.

  2. My friend, as an NDIS participant and ex lawyer, is extremely stressed at having to manage the legal contracts with her plan manager and her service providers. Many of these providers do not understand their legal obligations under the NDIS or the Consumer Law, the legal structures that they are using, or their employment law obligations to their own employees. My friend is left on her own to manage these issues as best she can, with no help from government.

4 Rachel at This Woman Votes - “American Epistemics: The Machine Doesn’t Need to Be Right,

Only Fast” https://twvme.substack.com/p/american-epistemics-the-machine-doesnt

5 Melissa Davey, “StuU of Nightmares: calls for help surge by 50% after Australia launches aged assistance tool” 6 See footnote 4.

4

  1. One recent example: she appears to have been overcharged by one service provider - with the approval of her plan manager.

  • The NDIS factsheet ‘working with providers’ says that ‘unregistered providers can choose to charge prices below or above the price limit’ (what incentive therefore is there to register?)

  • There is an exception to the usual pricing limits which seems to give providers the ability to charge night rates for their day work if they go even a few minutes over 8pm. But they are meant to ‘discuss’ this with the participant – which appears to mean that the participant’s consent is a precondition. This did not happen in the case of my friend. If one looks at the Facebook discussion groups of providers, it seems that everyone is confused about how to apply the exception. Another area ripe for fraud but which my friend is left to negotiate on her own.

Conclusion

The NDIS was a wonderful scheme in concept, but by establishing basically a privatised structure which is left to the disabled participants to manage, it has opened the gates for fraud. The answer is not to make the NDIS an AI- managed, automated system that continues to restrict participants and their services. The answer is to introduce the rule of law and human decision-making into the system. Funding legal assistance is one possible answer, but if the work of employing and regulating providers were to be managed by the government rather than NDIS participants, this would not be necessary. The Bill is not aimed in the right direction. Yours sincerely

Dr Tamsin Clarke