AI Inquiry Atlas

An unofficial, independent guide to the public records of the Joint Select Committee on Artificial Intelligence. Not connected with the Parliament of Australia or the committee. Its makers also made a submission (no. 481; see About). Summaries and scores are written by AI and can be wrong: check the original before relying on them.

Submission 522

Australian Nursing & Midwifery Federation (ANMF)

Submitter group: Unions.

An AI model identified the submitter group and wrote the summary, key points and recommendation summaries. Counts do not measure importance or merit.

Open in the interactive guide

Summary

Read the original submission for the full text.

Addresses AI in health and aged care from a professional and industrial perspective, treating AI as neither inherently beneficial nor harmful. Argues AI must support rather than replace human care and staffing, with worker consultation, sovereign data control, clear and enforceable regulation, and accountability that follows control.

Key points

Quotes are the submitter’s own words. Text taken from documents, including scanned pages read by text-recognition software, can differ from the PDF. Check the original. Listed in the same order as on the submission’s page in the interactive guide.

  1. Argues that AI must never justify reductions in safe staffing, nursing or midwifery positions, skill mix, ratios or care hours.

    AI must never be used to justify reductions in safe staffing, nursing or midwifery positions, skill mix, ratios or care hours.

  2. Argues that nurses, midwives, care workers and unions should be involved from problem definition and procurement through design, testing, implementation and review of AI systems.

    Nurses, midwives, care workers, and their unions should be involved from problem definition and procurement through design, testing, implementation, and review.

  3. Argues that regulation of AI should be clear, enforceable and coordinated, and should not wait for widespread adoption or a serious adverse event.

    The ANMF supports clear, enforceable, and coordinated regulation of AI. AI is already being used in health and care, and regulation cannot wait for widespread adoption or a serious adverse event before responsibilities are made clear.

  4. Argues that responsibility should not fall disproportionately on individual practitioners and accountability should follow control over the relevant decision.

    Complex or fragmented regulation must not push responsibility disproportionately onto individual practitioners. Accountability should sit with the people and organisations that have control over the relevant decision.

  5. Argues that Australian data storage alone does not establish sovereignty if overseas entities retain data access, control of infrastructure or models, or contractual control.

    storing data in Australia does not establish sovereignty where an overseas entity retains access to the data, controls the infrastructure or model, determines contractual terms, or can use Australian data to improve a proprietary product.

  6. Proposes an algorithm health equity assessment before high-impact health AI, including whether development and testing data adequately represent its intended population.

    We propose that organisations undertake an ‘algorithm health equity assessment’ before high-impact AI systems are introduced in health care. Such an assessment should examine whether development and testing data adequately represent the population in which the system will be used

Recommendations

Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.

  1. AI must never be used to justify reductions in safe staffing, nursing or midwifery positions, skill mix, ratios or care hours.

    AI must never be used to justify reductions in safe staffing, nursing or midwifery positions, skill mix, ratios or care hours.

  2. Adopt clear, enforceable and coordinated regulation of AI, without waiting for widespread adoption or a serious adverse event.

    The ANMF supports clear, enforceable, and coordinated regulation of AI. AI is already being used in health and care, and regulation cannot wait for widespread adoption or a serious adverse event before responsibilities are made clear.

  3. Do not rely on voluntary guidance alone for high-impact uses of AI in health and care.

    Voluntary guidance alone is not sufficient for high-impact uses of AI in health and care.

  4. Do not treat AI as a solution to workforce shortages or use it to mask or bridge chronic understaffing.

    AI must never be viewed as a solution to workforce shortages or used to mask or bridge chronic understaffing.

  5. Do not count AI efficiencies as nursing/midwifery hours or use them to alter staffing establishments, reduce rostered hours, dilute skill mix, undermine nurse-to-patient ratios or midwifery staffing models, or reduce mandated aged-care minutes.

    AI-generated efficiencies must not be counted as equivalent to nursing or midwifery hours or used to alter staffing establishments, reduce rostered hours, dilute skill mix, undermine nurse-to-patient ratios or midwifery staffing models, or reduce mandated care minutes in aged care.

  6. Keep human care available, including face-to-face assessment where required.

    Human care, including face-to-face assessment where required, must remain available.

  7. Involve nurses, midwives and care workers in defining problems, planning, procurement and testing before any vendors are confirmed.

    Nurses, midwives and carers who will use these systems should be involved in problem identification, planning and procurement, testing and decision-making before any vendors are confirmed.

  8. Evaluate AI in intended clinical settings and workflows, using staged implementation, simulation or formative evaluation where appropriate, and monitor after implementation.

    Systems should be evaluated in the clinical settings and workflows in which they will be used, including through staged implementation, simulation or other formative evaluation where appropriate, and monitored after implementation.

  9. Where AI supports clinical decisions, nurses and midwives must retain a genuine ability to question, disregard, override or escalate an output.

    Where AI supports clinical decisions, nurses and midwives must retain a genuine ability to question, disregard, override, or escalate an output.

  10. Australian sovereign AI capability should include the capacity to develop, host, inspect, govern, adapt and, where necessary, replace the AI systems on which essential public services depend.

    Australian sovereign AI capability should include the capacity to develop, host, inspect, govern, adapt and, where necessary, replace the AI systems on which essential public services depend.

  11. Invest in smaller and specialised AI systems suited to Australian economic, cultural and public service needs rather than reproducing the largest overseas models.

    Australia does not need to reproduce the largest overseas models to build useful sovereign capability. There is a strong case for investment in smaller and specialised systems suited to Australian economic, cultural, and public service needs.

  12. Public investment in AI capability should deliver enduring public benefit, including secure jobs, workforce development, domestic computing and research infrastructure access, local supply chains and retention of IP and economic value in Australia.

    Public investment in AI capability should deliver enduring public benefit, including secure jobs, workforce development, access to domestic computing and research infrastructure, local supply chains, and the retention of intellectual property and economic value in Australia.

  13. Protect patient, workforce and organisational data throughout the AI lifecycle and ownership or outsourcing changes, covering secondary use, training, fine-tuning, subcontracting and workable exits.

    Patient, workforce and organisational data must be protected throughout the lifecycle of AI technologies, including where products are transferred, acquired, outsourced or managed by another owner. Protections should cover secondary use of data, model training and fine- tuning, ownership changes, subcontracting, and workable exit arrangements.

  14. Extend data protections to information generated through AI use, such as prompts, outputs, logs, embeddings, inferred information and model telemetry.

    These protections must extend to information generated through AI use, such as prompts, outputs, logs, embeddings, inferred information and model telemetry, not only the original dataset supplied to a system.

  15. Do not transfer or make available sensitive clinical and workforce data for AI training without clear lawful authority, strict governance and demonstrable public benefit.

    Sensitive clinical and workforce data should not be transferred or made available for AI training without clear lawful authority, strict governance, and demonstrable public benefit.

  16. For Australian public data used to develop or improve AI, define purpose and duration, preserve public control, prevent unauthorised proprietary-model incorporation and deliver clear public benefit.

    Where Australian public data is used to develop or improve AI, its use should have a defined purpose and duration, preserve public control, prevent unauthorised incorp oration into proprietary models and deliver a clear public benefit.

  17. For AI using Aboriginal and Torres Strait Islander data, involve their communities, Aboriginal Community Controlled Health Organisations, Aboriginal health workers and practitioners, and Aboriginal nurses and midwives as relevant in governance and evaluation.

    Where AI systems use data relating to Aboriginal and Torres Strait Islander peoples, governance and evaluation should involve Aboriginal and Torres Strait Islander communities, Aboriginal Community Controlled Health Organisations, Aboriginal health workers and practitioners, and Aboriginal nurses and midwives as relevant.

  18. Assess systems for cultural safety so they do not reinforce deficit-based assumptions or reduce access to culturally safe, community-led care.

    Systems should also be assessed for cultural safety and should not reinforce deficit-based assumptions or reduce access to culturally safe, community-led care.

  19. Test health AI systems for the populations and settings in which they will be used, with nurses, midwives and affected communities involved in judging whether they are safe and useful.

    Systems still need to be tested for the populations and settings in which they will be used, with nurses, midwives , and affected communities involved in judging whether they are safe and useful.

  20. Do not use AI to reduce nursing or midwifery positions, weaken staffing or skill mix, or reclassify workers in ways that diminish standing, responsibilities, conditions or pay.

    AI should improve jobs, not remove the workforce from health and care. It should not be used to reduce nursing or midwifery positions, weaken staffing or skill mix, or reclassify workers in ways that diminish professional standing, responsibilities, conditions, or pay.

  21. Assess the workload effects of AI before implementation and evaluate them after, including time for checking, escalation, training, workarounds and system failure.

    The workload effects of AI should therefore be assessed before implementation and evaluated after implementation, including the time required for checking, escalation, training, workarounds, and system failure.

  22. Do not treat an efficiency as a workforce saving unless it can be shown to reduce unnecessary work without reducing care, increasing risk or shifting work elsewhere.

    In health care, a claimed efficiency should not be treated as a workforce saving unless it can be shown to reduce unnecessary work without reducing care, increasing risk, or shifting work elsewhere.

  23. Involve nurses, midwives, care workers and their unions from problem definition and procurement through design, testing, implementation and review, with iterative consultation.

    Nurses, midwives, care workers, and their unions should be involved from problem definition and procurement through design, testing, implementation, and review. Consultation should be iterative because the effects of AI on work may only become clear once a system is used in practice.

  24. Give workers enforceable rights to information, consultation, paid training and retraining, review of AI-influenced decisions and union representation.

    Workers should have enforceable rights to information, consultation, paid training and retraining, review of AI-influenced decisions and union representation.

  25. Do not use AI to analyse performance, discipline, recruit, roster or allocate work without transparency, meaningful human review and a clear way for workers to challenge a decision.

    AI should not be used to analyse performance, discipline, recruit, roster or allocate work without transparency, meaningful human review and a clear way for affected workers to challenge a decision.

  26. Place limits on the collection and use of worker data.

    There must also be limits placed on the collection and use of worker data.

  27. Workers should always know when AI or surveillance technologies are used, how they affect decisions about work, and how decisions can be questioned or challenged.

    Workers should always know when AI or surveillance technologies are being used, how they affect decisions about work, and how those decisions can be questioned or challenged.

  28. Build practical AI risk literacy in education so nurses, midwives and care workers can recognise risks and limitations in AI outputs and know when to question, stop or escalate its use.

    Education should build practical AI risk literacy so nurses, midwives and care workers can recognise risks and limitations in AI outputs and know when to question, stop or escalate its use.

  29. Provide ongoing, accessible training and implementation support in paid work time.

    Training and implementation support should be ongoing, accessible, and provided in paid work time.

  30. Build specialist nursing and midwifery informatics, education, research and digital leadership capability in the health system.

    The health system also needs specialist nursing and midwifery informatics, education, research, and digital leadership capability.

  31. Have experts who understand the IT system and its healthcare environment deliver nursing and midwifery AI education; where possible use nurse informatics experts.

    Education for nurses and midwives should be delivered by subject matter experts that understand both the IT system and the environment it is intended to function in (including healthcare). For example, where possible, training should be delivered by the nurse informatics experts described above.

  32. Address AI capability in pre-registration nursing and midwifery education.

    AI capability must also be addressed in pre-registration nursing and midwifery education.

  33. Do not use AI to reduce access to clinical educators, preceptors, supervision or supported transition to practice.

    AI must not be used to reduce access to clinical educators, preceptors, supervision or supported transition to practice.

  34. Set high Commonwealth AI standards, especially in health, aged care and community services, covering procurement, privacy, security, governance, early worker consultation, education and ongoing service and workforce monitoring.

    The Commonwealth should set a high standard in its own use of AI, especially in health, aged care and other services delivered to the community. This includes appropriate procurement, privacy, security and governance, as well as early workforce consultation, staff education, and ongoing monitoring of how AI affects services and work.

  35. Implement the commitments to foundational training, staff consultation and the GovAI platform in ways that let workers and unions influence system design and provide transparent evidence on service quality, workload and employment.

    It commits to foundational training, staff consultation and engagement, and GovAI as an Australian -based, vendor-agnostic platform intended to provide secure access to multiple models and reduce vendor lock-in.9 These commitments should be implemented in ways that allow workers and their unions to influence system design and provide transparent evidence about effects on service quality, workload and employment.

  36. Explain AI use, purpose, limitations, checks, questioning and override options, and escalation in plain-language worker guidance.

    Worker-facing guidance should explain, in plain language, whether AI is being used, what the system is intended to do, its important limitations, what the worker is expected to check, when an output can be questioned or overridden, and how concerns should be escalated.

  37. Make employers and organisations responsible for system due diligence before implementation, worker consultation throughout and evaluation afterwards, including data storage, appropriate training data and safeguards.

    The burden of establishing whether an approved workplace system has appropriate data- storage arrangements, context-appropriate training data, or adequate technical safeguards should not fall exclusively on the shoulders of frontline workers . Rather, these are organisational and employer responsibilities requiring due diligence before implementation, consultation with impacted workers throughout and ongoing evaluation thereafter.

  38. Build data portability, interoperability, open standards, auditable access arrangements and workable exit provisions into Commonwealth procurement contracts.

    Commonwealth procurement should also protect service continuity and public control. Data portability, interoperability, open standards, auditable access arrangements and workable exit provisions should be built into contracts so agencies can change provide rs without losing access to data or disrupting essential services.

  39. Consider ACTU model licensing for high-impact and general-purpose AI, plus healthcare-specific regulation: proactive developer/deployer risk mitigation, ongoing transparency, regular safety audits, high-risk guardrails, governance and major-incident reporting, backed by meaningful non-compliance penalties.

    The ANMF supports the call made by Australian Council of Trade Unions (ACTU) Secretary Melissa Donnelly for a licensing regime for AI models that would require proactive risk mitigation by developers and deployers, ongoing transparency, regular safety auditing, guardrails against high-risk use, effective governance and mandatory reporting of major incidents, backed by meaningful penalties for non-compliance.10 An approach such as this warrants serious consideration, both for high-impact and general-purpose AI systems, as well as sector-specific regulation in industries like healthcare.

  40. Have Ahpra, the Nursing and Midwifery Board, the Commission and other regulators work with nurses, midwives and their representatives to provide clearer profession-specific guidance on AI.

    Ahpra, the Nursing and Midwifery Board of Australia, the Commission and other relevant regulators should work with nurses, midwives and their representatives to provide clearer profession-specific guidance as AI use develops.

  41. Actively evaluate early-use settings such as primary care, aged care and disability services while systems and models of care can still be adapted safely.

    Regulation and assurance should also examine settings where AI-enabled technologies are already being used. Primary care, aged care and disability services provide examples where AI tools, virtual care and automated assessment systems are already entering practice, possibly prior to the existence of rigorous mechanisms to ensure their appropriateness. ANMF members have identified examples where algorithms may restrict professional judgement or provide no meaningful ability to override an outcome. Early-use settings should therefore be actively evaluated while systems and models of care can still be adapted safely.

  42. Do not push responsibility disproportionately onto individual practitioners; place accountability with the people and organisations controlling the relevant decision.

    Complex or fragmented regulation must not push responsibility disproportionately onto individual practitioners. Accountability should sit with the people and organisations that have control over the relevant decision.

  43. Ensure nurses and midwives have enough information, time and authority to question, disregard or escalate an AI output.

    Nurses and midwives must have enough information, time and authority to question, disregard or escalate an AI output.

  44. Allow staff to report concerns without reprisal and provide clear escalation pathways when AI outputs conflict with clinical judgement.

    Staff should be able to report concerns without reprisal, and there should be clear escalation pathways when AI outputs conflict with clinical judgement.

  45. Where AI is already used in health and care, have regulators proactively identify and close gaps now.

    Where AI is already being used in health and care, regulators should proactively identify and close gaps now.

  46. AI systems in health care should support rather than replace human practitioners.

    Within health care, AI systems should support rather than replace human practitioners.

  47. Retain human or non-AI pathways where clinically required or better suited to a person's circumstances, preferences or needs; do not force AI pathways because they are cheaper.

    People should not be forced into AI or automated pathways because they are cheaper. Human or non-AI pathways must remain available where they are clinically required or better suited to a person’s circumstances, preferences or needs.

  48. Introduce regulated protections against the misleading use of protected professional titles such as 'nurse' and 'midwife' by AI chatbots or automated systems where no registered professional provides the service.

    There must be regulated protections against the misleading use of protected professional titles such as ‘nurse’ and ‘midwife’ for AI chatbots or other automated systems where a registered professional is not providing the service.

  49. Give health and care workers avenues to report deepfakes, harassment and intrusive surveillance, seek removal or correction where possible, and obtain workplace and regulatory support.

    AI-enabled harms can also be workplace harms. Donnelly (2026) has identified deepfakes, harassment and intrusive surveillance among harms already experienced by workers. 10 Health and care workers should have clear avenues to report such harms, have material removed or corrected where possible, and obtain workplace and regulatory support.

  50. Test AI systems for the populations and clinical settings where they will actually be used, including populations historically poorly represented in health research and data.

    Systems need to be tested for the populations and clinical settings in which they will actually be used, including populations historically poorly represented in health research and data.

  51. Require independent validation in the intended clinical setting and continued evaluation after implementation.

    This example demonstrates why independent validation in the intended clinical setting, and continued evaluation after implementation, are necessary even where a system is already widely used.

  52. Propose equity assessments before introducing high-impact health AI, examining data representation, group performance, unfair access or resource effects, and required safeguards, clinical oversight and monitoring.

    We propose that organisations undertake an ‘algorithm health equity assessment’ before high-impact AI systems are introduced in health care. Such an assessment should examine whether development and testing data adequately represent the population in which the system will be used, whether performance differs between population groups, whether use of the system could change access to care or resources unfairly, and what safeguards, clinical oversight and monitoring are required.

  53. Explicitly consider cultural safety, particularly for Aboriginal and Torres Strait Islander peoples and other populations poorly represented in datasets or affected by historical bias in health data.

    Cultural safety should be considered explicitly, with particular attention to Aboriginal and Torres Strait Islander peoples and other populations who may be poorly represented in existing datasets or affected by historical bias in health data.

  54. Treat data portability, interoperability, open standards, clear subcontracting arrangements, Australian control over sensitive information and tested exit and business-continuity arrangements as part of critical infrastructure planning.

    Data portability, interoperability, open standards, clear subcontracting arrangements, Australian control over sensitive information, and tested exit and business-continuity arrangements should be treated as part of critical infrastructure planning.

  55. Resource Ahpra, the Nursing and Midwifery Board, ANMAC and the TGA with specialist expertise and dedicated staff to monitor AI, prepare the workforce and assess contributions to poor clinical or professional outcomes.

    Regulators such as Ahpra, the Nursing and Midwifery Board of Australia, the Australian Nursing and Midwifery Accreditation Council and the Therapeutic Goods Administration must be resourced with specialist expertise and dedicated staff to monitor AI use, prepare the workforce and assess and the contribution of AI to poor clinical or professional outcomes.

  56. Ensure regulatory capability includes nursing, midwifery, clinical and workforce expertise, not technical and digital health expertise alone.

    Regulatory capability must include nursing, midwifery, clinical and workforce expertise, not technical and digital health expertise alone, where historically much of the decision-making to date has sat.

  57. Consider whether the current distribution of powers is sufficient for high-impact and general-purpose AI.

    The Committee should also consider whether the current distribution of powers is sufficient for high-impact and general-purpose AI.

  58. Support the ACTU proposal for a strengthened AI Safety Institute to administer a model-licensing regime.

    We support the ACTU proposal calling for a strengthened AI Safety Institute administer a model-licensing regime.

  59. Whatever institutional model is adopted, link technical testing and incident detection to enforceable regulatory action while retaining health, professional, workplace and other specialist regulators' responsibilities.

    Whatever institutional model is adopted, there must be a clear pathway from technical testing and incident detection to enforceable regulatory action, while retaining the responsibilities of health, professional, workplace and other specialist regulators.

  60. Judge health AI productivity by safer, more accessible therapeutic care, reduced unnecessary work and clinicians' capacity; augment documentation and handover rather than automate them away.

    In health, productivity should be judged by whether care is safer, more accessible and therapeutic, whether unnecessary work is reduced, and whether nurses, midwives and care workers have more time and capacity to provide high-quality care. Time returned to care includes assessment, clinical reasoning, communication, care planning, clinical supervision and professional reflection and safe handover. Documentation and clinical handover can themselves be part of clinical care, supporting communication, reflection, informal coaching opportunities and care planning, AI should augment these practices rather than treat them as low-value tasks than can be automated out of clinician’s hands.

  61. Use AI to strengthen local health services and care workers in regional, rural and remote communities.

    AI has the potential to improve access to health care in regional, rural, and remote communities. It may improve access to specialist knowledge, support clinical decision-making and diagnostics, enable remote monitoring, and provide new opportunities for education and professional support. These benefits are particularly relevant where distance and limited access to specialist services already affect care. AI should be used to strengthen the capacity of local health services and the nurses, midwives, and care workers who provide them.

  62. Do not use automated or virtual options to lower regional, rural or remote care standards or replace access to human practitioners to save costs.

    People in regional, rural and remote communities should not receive a lower standard of care because an automated or virtual option is available, and these systems should not replace a person's access to human practitioners as a means of cost-saving.

  63. Ensure AI research and innovation addresses problems that matter to care recipients and workers.

    AI research and innovation in healthcare should address problems that matter to people receiving care and the workforce.

  64. Retain jurisdictional and operational control over data access, processing, transfer and repurposing, including secondary use, training, fine-tuning, ownership changes, subcontracting and exit arrangements.

    Effective sovereignty requires jurisdictional and operational control over how information is accessed, processed, transferred and repurposed, including secondary use, model training and fine-tuning, ownership changes, subcontracting and workable exit arrangements.

  65. Do not convert publicly generated data into private products that government must later license back.

    Publicly generated data should not simply be converted into privately controlled products that governments must later licence back from external providers.

  66. Provide role-specific guidance on AI use, purpose, limitations, checks, stopping, override, escalation, permitted inputs and responsibility.

    They do need practical, role-specific guidance about when AI is being used, what it is intended to do, its known limitations, what must be checked, when to stop, override or escalate, what information can be entered and where responsibility sits.

  67. Have education providers and accreditation bodies such as ANMAC consider AI's effects on curriculum, assessment and preparation for practice, and support educators' professional development.

    Education providers and accreditation bodies such as the Australian Nursing and Midwifery Accreditation Council should consider how AI affects curriculum, assessment and preparation for practice, while educators need appropriate support and professional development.

Read the original

Open in the interactive guide