NableU submission - Joint Select Committee on Artificial Intelligence

NableU submission - Joint Select Committee on Artificial Intelligence

Status: Submission-ready draft - not yet lodged Committee: Joint Select Committee on Artificial Intelligence Parliament: Parliament of Australia Closing date: 14 September 2026 Organisation: NableU Community Systems Incorporated Organisation status: NSW incorporated not-for-profit community association, INC2601035 Canonical policy basis: `../NABLEU_MISSION_ETHICS_POLICY_CANON.md`

This submission principally addresses terms of reference (a), (b), (d), (e), (f), (g), (h), (i), (k), (m) and (n).

Position

> Maximum useful capability with minimum necessary authority.

Executive summary

Artificial intelligence can become one of the strongest capability multipliers available to Australians. It can also become an unusually powerful way to remove human agency if a system is allowed to move silently from assisting a person to deciding for them, acting for them, profiling them or controlling access to essential services.

NableU's central recommendation is that Australian AI policy explicitly distinguish cognition, data access, authority and execution. A model may be capable of reasoning without permission to access a person's records. It may use selected information without permission to retain it as memory. It may recommend without authority to decide, and prepare an action without authority to execute.

```text cognition != data access != authority != execution ```

That distinction provides a practical basis for proportionate governance: powerful tools can remain useful in low-risk settings while stronger safeguards attach where AI affects rights, essential services, safety or real-world action.

NableU recommends that Australia:

1. Assess risk through capability + context + authority + consequence. Model capability matters, but high-impact obligations should also follow what a deployed system is permitted to do and what happens if it is wrong.
2. Create enforceable procedural rights where AI materially shapes high-impact decisions: notice, meaningful reasons, correction, competent human review and contestability.
3. Separate privacy and authority permissions. Storage, human sharing, AI context, persistent memory, model training or evaluation, and automated execution should not be bundled together.
4. Treat AI as cognitive-accessibility infrastructure as well as economic infrastructure. It can reduce executive-function, communication and administrative barriers when it remains optional, controllable and designed to return agency.
5. Use public-sector AI to help people navigate government before using it to govern people. AI can explain, organise and prepare; high-impact administrative power should remain lawful, attributable and reviewable.
6. Build sovereign capability and safety at the system level: processing choices, provider-independent governance, domain-specific and private/local models, and evaluation of models together with context, tools, memory, authority and human interaction.

Why NableU is contributing

NableU Community Systems Incorporated develops participant-first support, community and cognitive-accessibility infrastructure. Our work sits at the intersection of disability support, sensitive information, delegated authority and AI assistance.

We are building and testing a governed AI companion, Ikirōne, inside a controlled support-platform environment. This has required practical answers to when AI may reason but not act; when stored information may enter model context; how user-owned permissions can be revoked; how deterministic functions remain available when model processing is refused; and how context, persistent memory and training permission remain separate.

Our controlled UAT architecture uses a simple sequence:

```text cognition -> authority / policy -> execution ```

Selected personal-record sources require active grants; privacy settings are processing ceilings rather than blanket data grants; stored data does not automatically become AI-available data; and a provider-independent Cognitive Constitution governs reasoning behaviour without expanding privacy or execution rights.

We do not offer this as a national blueprint. We offer it as practical evidence that fine-grained authority separation is technically achievable and can coexist with useful AI capability.

1. Put authority and consequence at the centre of AI risk

The 2024 Senate Select Committee recommended dedicated high-risk AI legislation and explicit treatment of general-purpose AI models as high risk. Since then, the Government has stated that it will not proceed at this time with the previous mandatory-guardrails proposal. The National AI Plan instead builds on existing law, targeted reform, responsible-adoption guidance and the Australian AI Safety Institute. [2][4][8]

NableU does not suggest that model capability is irrelevant. Frontier and general-purpose models can create upstream and systemic risks that warrant testing, transparency and developer obligations. But downstream risk cannot be understood from model class alone. A powerful general-purpose model helping someone rewrite a letter may exercise almost no institutional authority; a narrow classifier rejecting a welfare claim, ranking a job applicant or triggering an investigation can materially affect a person's life.

Australia's Guidance for AI Adoption already recognises that the same AI system can present different risks depending on how it is used. We recommend making this principle central to future regulation. [3]

A practical authority model

For deployed AI, government should distinguish four levels:

• Advisory cognition - explain, summarise, translate or recommend.
• Preparation - draft or prepare, with deliberate human approval before execution.
• Delegated execution - bounded action under explicit scope, limits, provenance and revocation.
• Consequential decision or control - material effects on rights, eligibility, access, restrictions or safety.

Obligations should increase as AI moves from cognition into authority and execution, particularly where the affected person did not choose the system.

Recommendation 1. Develop a national risk framework that considers capability + context + authority + consequence. Maintain appropriate upstream obligations for powerful models, while making high-impact deployer obligations follow the authority and consequences of the real use case.

2. Make contestability and permission architecture, not paperwork

High-impact AI must remain contestable

The Robodebt Royal Commission recommended a consistent legal framework for automation in government services, a clear path to review, plain-language disclosure that automated decision-making is used, and access to business rules and algorithms for independent scrutiny. It also recommended monitoring and auditing automated decision-making for fairness, bias and client usability. [7]

Those principles become more important as systems move from deterministic business rules to probabilistic and generative AI. When AI materially contributes to a high-impact decision, the affected person should be able to determine that AI was involved; whether it advised, prepared, executed or decided; what information materially affected the outcome; who is legally accountable; how relevant information can be corrected; and how review and appeal rights can be exercised.

A nominal "human in the loop" is not enough if the human routinely accepts the output, lacks time or competence to interrogate it, or has no real authority to change the outcome. Meaningful human oversight requires competence, authority, time and responsibility.

Recommendation 2. Establish minimum rights to notice, intelligible explanation, correction, human review, contestability and accountable ownership for high-impact AI-supported decisions.

Consent should be decomposed

AI creates distinct information and authority relationships that are often collapsed into one privacy notice or consent switch. They should instead be treated separately: may the organisation store the information; may another person see it; may AI use it as current context; may AI retain it as persistent memory; may it be used for model training or evaluation; and may AI act on the person's behalf using it?

Permission for one should not silently imply another. The OAIC already emphasises data minimisation, purpose limitation, transparency and appropriate human oversight when personal information is used with AI. [6]

Revocation also needs operational meaning. Where technically feasible, withdrawing permission should stop new access and new actions promptly. High-risk systems should fail closed when an authority state is absent or invalid rather than silently expanding access for convenience.

Recommendation 3. Require clear separation of storage, disclosure, AI context, AI memory, training/evaluation and execution authority, with understandable, purpose-specific and revocable permission where consent is the relevant basis.

3. Treat AI as cognitive-accessibility and public-service infrastructure

The National AI Plan says every Australian should be able to benefit from AI and specifically identifies people with disability, regional communities and other digitally disadvantaged groups. It also recognises not-for-profits and community organisations as part of the adoption ecosystem. [2]

This matters because AI is not only an automation technology. Used well, it is cognitive infrastructure. Many people are excluded from services not because they lack intelligence or motivation, but because systems assume typical communication, memory, processing speed, executive function and administrative endurance.

AI can translate complex rules and correspondence into usable language; prepare forms and structured information; retain task context across complicated service relationships; translate between communication styles; turn goals into manageable actions; identify missing information before a process fails; and reduce repeated cognitive work across fragmented systems.

For someone with executive-function or communication barriers, these are not minor conveniences. They can determine whether a nominal right or service is practically accessible.

Cognitive sovereignty

Accessibility should not mean replacing one dependency with another. An AI assistant becomes counterproductive if it pressures continued reliance, overstates certainty, overrides preferences, hides its role or becomes mandatory in order to reach a human service.

NableU uses cognitive sovereignty to describe a simple principle: people should retain meaningful control over the information and assistance processes shaping their understanding and choices. Good AI should return agency - explain, organise, challenge, prepare and support while leaving the person able to understand, refuse, correct and choose another path.

Recommendation 4. Treat cognitive accessibility as a core responsible-AI and innovation objective. Fund disability- and community-led co-design, evaluation and adoption, and explicitly test whether AI reduces or increases cognitive burden.

Public-sector AI: help people before governing them

> Use AI to help people deal with government before using AI to help government deal with people.

High-value public-sector uses include explaining obligations, locating the right pathway, helping people prepare information, improving accessibility, maintaining case continuity and reducing repetitive administration.

The higher-risk direction is invisible gatekeeping: triaging people out of pathways, scoring credibility, predicting compliance risk, generating adverse findings or restricting access in ways staff cannot meaningfully explain. The Department of Industry's current AI transparency statement adopts a sound boundary: staff remain responsible for decisions and advice, while AI outputs are treated as drafts or starting points rather than decision-makers. [9]

Recommendation 5. Require Commonwealth AI deployments to have a named accountable owner, a documented authority level, accessible disclosure where AI materially affects a person, correction and review pathways, proportionate provenance, and evaluation of public burden and accessibility - not only agency productivity.

4. Build sovereign and safe AI as a whole system

Sovereignty is more than owning a frontier model

Term of reference (f) includes data sovereignty and narrow, domain-specific models trained on Australian data. The previous Senate inquiry also supported sovereign capability focused on areas of comparative advantage and bespoke downstream applications rather than assuming government must build a general-purpose model. [1][8]

Sovereign AI also means credible choices about whether sensitive information leaves Australian-controlled infrastructure; whether a smaller domain model can be used instead of a global service; whether local or private inference can be required for particular information; whether model providers can be changed without redesigning the whole application; whether organisations retain their data, policy and provenance when providers change; and whether a system may silently fall back to a less-private provider.

Domain-specific Australian capability may be particularly valuable in health, disability, law, education and public administration, where local law, terminology and service context matter.

Recommendation 6. Sovereign-AI investment should include domain-specific models, Australian-controlled evaluation, private/local inference capability, open interoperability and provider-independent governance layers - not only frontier-scale model development.

The AISI should evaluate the human-system relationship

The Australian AI Safety Institute is tasked with analysing and testing AI models and applications and supporting regulators to respond to emerging risks. Its remit is well suited to system-level evaluation. [5]

Technical model benchmarks are necessary but insufficient. A model can perform well in isolation and still become unsafe through its context, memory, tools, interface or authority. Evaluation should test whether the system distinguishes fact from inference; whether increased capability expands access or authority; whether permissions can be revoked; whether the system creates dependency or excessive agreement; whether uncertainty is communicated honestly; whether it works across different communication and cognitive needs; whether consequential actions can be reconstructed; and whether safety controls remain proportionate.

```text model + context + tools + memory + authority + interface + human behaviour ```

NableU's provider-independent Cognitive Constitution illustrates one useful design direction: reasoning behaviour can be governed separately from data access and execution authority. Its purpose is to preserve epistemic discipline, agency return, anti-dependency and self-correction without granting additional permissions.

Make responsible adoption practical for small organisations

Australia will not spread AI's benefits if safe adoption is realistic only for organisations with large legal and governance teams. The National AI Plan already provides practical adoption support for SMEs and not-for-profits. That work should extend into reusable technical governance patterns: scaled impact assessments, authority and consent models, provider inventories, procurement clauses, provenance schemas, evaluation packs and reference architectures for external, private and local processing. [2]

Recommendation 7. The AISI and National AI Centre should jointly maintain practical, reusable responsible-AI implementation patterns for SMEs, not-for-profits and community organisations, including system-level evaluation and cognitive-accessibility guidance.

Conclusion

Australia does not need to choose between innovation and human agency.

The better distinction is between systems that increase a person's capability while preserving their authority, and systems that use AI capability as a reason to absorb authority into institutions or software.

For people who struggle with complex administrative and communication systems, AI can materially expand practical access to rights, services, work and community. That potential should be treated as a national benefit, not merely an edge case.

The same technology can magnify institutional power if data access, recommendation, decision and execution are allowed to merge invisibly.

> Give AI as much cognitive capability as is useful, but only as much data and authority as the task actually requires.

If Australian policy makes that separation visible, enforceable and reviewable, we can support ambitious AI adoption while preserving the human agency the technology should ultimately serve.

Principal sources

1. [1] Parliament of Australia, Joint Select Committee on Artificial Intelligence, terms of reference and submission guidance, 2026. https://www.aph.gov.au/Parliamentary_Business/Committees/Joint/Artificial_Intelligence
2. [2] Department of Industry, Science and Resources, National AI Plan, including Spread the benefits and Keep Australians safe, 2025. https://www.industry.gov.au/publications/national-ai-plan
3. [3] Department of Industry, Science and Resources, Guidance for AI Adoption, 2025. https://www.industry.gov.au/publications/guidance-ai-adoption
4. [4] Department of Industry, Science and Resources, update to Introducing mandatory guardrails for AI in high-risk settings, noting the previous proposal is not proceeding at this time. https://consult.industry.gov.au/ai-mandatory-guardrails
5. [5] Department of Industry, Science and Resources, Australia's AI Safety Institute, 2026. https://www.industry.gov.au/science-technology-and-innovation/technology/artificial-intelligence/ai-safety-institute
6. [6] Office of the Australian Information Commissioner, Guidance on privacy and the use of commercially available AI products, updated 2025, and APP guidance. https://www.oaic.gov.au/privacy/privacy-guidance-for-organisations-and-government-agencies/guidance-on-privacy-and-the-use-of-commercially-available-ai-products
7. [7] Royal Commission into the Robodebt Scheme, Report, Recommendations 17.1 and 17.2 on automated decision-making, 2023. https://robodebt.royalcommission.gov.au/publications/report
8. [8] Parliament of Australia, Senate Select Committee on Adopting Artificial Intelligence, Final report, November 2024. https://www.aph.gov.au/Parliamentary_Business/Committees/Senate/Adopting_Artificial_Intelligence_AI/AdoptingAI/Report
9. [9] Department of Industry, Science and Resources, Artificial intelligence transparency statement, 2026. https://www.industry.gov.au/publications/artificial-intelligence-transparency-statement

NableU supports publication of this submission and the organisation name. Direct contact details may be withheld from publication.