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 534

Community and Public Sector Union (CPSU)

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.

Argues from the perspective of Australian Public Service workers that AI adoption should be worker-led, share productivity gains and be subject to enforceable transparency, human accountability and independent auditing. Also addresses sovereign capability, data sovereignty, data centre regulation, workplace rights, copyright and an expanded, empowered AI Safety Institute.

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 workers must benefit from AI productivity gains, for example through better pay and conditions, reduced working hours and improved job quality.

    Workers must benefit from the productivity gains of their work and adoption of AI should not be at their expense. This includes sharing productivity gains through measures such as better pay and conditions, reduced working hours and improved job quality.

  2. Argues that government should be cautious about forecasting large public sector savings from AI and must factor in implementation, oversight and rework costs.

    Government should exercise caution in forecasting, or assuming, large-scale public sector savings. The cost of implementation, oversight and monitoring, and re-work must be factored into efficiency gains assessments, as well as increased case escalation and case complexity.

  3. Argues that stronger central oversight, enforcement powers and independent auditing are needed to ensure consistent APS standards.

    Stronger central oversight and enforcement powers are needed, supported by independent auditing, to ensure consistent standards across the APS.

  4. Argues that workers should have a statutory right to pre-decision consultation on AI, including through amendment of the Fair Work Act.

    Before decisions are made on AI-related workplace change, workers must have a right to consultation through Enterprise Agreements and the Fair Work Act 2009. This includes amending the Fair Work Act 2009 to provide a statutory right to pre-decision consultation on AI.

  5. Argues that sovereign AI capability can improve resilience by reducing reliance on multinational vendors and can increase transparency and accountability of AI.

    Developing sovereign AI capability can improve Australia's national and strategic resilience by reducing reliance on multinational vendors. In addition, it can increase transparency and accountability of AI by providing the access needed to understand, scrutinise and govern AI models.

  6. Recommends investigating a licensing scheme for frontier AI models, managed by the AI Safety Institute, requiring AI developers and deployers to be licensed to deploy products into Australia.

    Australian unions also recommend investigating a licensing scheme for AI frontier models, requiring AI developers and deployers to be licensed to deploy their products into Australia, to be managed by the AISI.

Recommendations

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

  1. Measure public-sector AI adoption by better citizen outcomes and working lives, and centre workers in its design, deployment and use.

    The successful adoption of Artificial Intelligence (AI) in the public sector should be measured by whether it delivers better outcomes for the citizens who rely on those services and improves the working lives of public servants. Public sector workers, with their deep understanding of government systems and service delivery, are best placed to lead AI implementation. Worker voices must be centred in the adoption of AI, with workers shaping how the technology is designed, deployed and used.

  2. Share AI productivity gains with workers through better pay and conditions, reduced working hours and improved job quality; do not adopt AI at workers’ expense.

    Workers must benefit from the productivity gains of their work and adoption of AI should not be at their expense. This includes sharing productivity gains through measures such as better pay and conditions, reduced working hours and improved job quality.

  3. Exercise caution in forecasts of public-sector AI savings; count implementation, oversight, monitoring, rework, increased case escalation and case complexity, and protect lawful, accurate decisions and reliable services.

    Government should exercise caution in forecasting, or assuming, large-scale public sector savings. The cost of implementation, oversight and monitoring, and re-work must be factored into efficiency gains assessments, as well as increased case escalation and case complexity. A focus on efficiency must not come at the expense of lawful, accurate decision making and reliable public services, as demonstrated by the lessons of Robodebt.

  4. Include rising-token-cost scenarios in AI cost modelling.

    Where AI adoption increases, token usage and associated costs may increase rapidly and modelling must include scenarios where token costs continue to rise.

  5. Have AI infrastructure support domestic capability rather than send profits offshore, particularly to multinational providers that do not pay their fair share of Australian tax.

    AI infrastructure must support domestic capability and not drive profits offshore to multinational providers, particularly those who do not pay their fair share of tax in Australia.

  6. Actively redistribute AI wealth so suburbs and regions share benefits rather than concentrating them among technology owners, investors and infrastructure-hosting regions.

    AI-driven wealth will require active redistribution to ensure the benefits are shared across Australia's suburbs and regions and not concentrated among technology owners, investors and the regions that host AI infrastructure.

  7. Ensure government procurement policies help share the economic benefits of AI across Australia, so companies benefiting from government contracts contribute fairly to Australia's tax base.

    Government procurement policies should help ensure the economic benefits of AI are shared across Australia. This should extend beyond improved public service delivery to ensure that companies benefiting from government contracts contribute fairly to Australia's tax base.

  8. Strengthen Statement of Tax Record requirements under the Shadow Economy Procurement Connected Policy so foreign-owned technology suppliers receiving government contracts demonstrate genuine tax compliance.

    The Government should strengthen the Statement of Tax Record requirements under the Shadow Economy Procurement Connected Policy to ensure foreign-owned technology suppliers receiving government contracts demonstrate genuine tax compliance.

  9. Require suppliers to certify tax-law compliance; no Australian audit or dispute in five years; no known circumstances likely to prompt Australian or overseas audits, investigations or disputes; no arrangements contravening anti-avoidance, transfer-pricing or diverted-profits tax provisions; no return or tax-liability extensions; and correct employee/contractor classification.

    At a minimum, suppliers should certify they:  have complied with all applicable tax laws,  have not been subject to a tax audit or dispute in Australia in the previous five years,  are not aware of circumstances likely to give rise to a tax audit, investigation or dispute in Australia or overseas,  have not engaged in arrangements that contravene anti-avoidance, transfer pricing or diverted profits tax provisions,  are not benefiting from extensions of time to lodge tax returns or resolve tax liabilities, and  appropriately classify workers and do not misclassify employees as contractors.

  10. Consider establishing data trusts as a mechanism for collective data rights.

    One option is the establishment of data trusts, which could provide a mechanism for collective data rights

  11. Consider governing data trusts through stakeholder representation, fiduciary obligations and a focus on benefits to members.

    The governance of data trusts could draw on principles that underpin Australia's industry superannuation system, including stakeholder representation, fiduciary obligations and a focus on delivering benefits to members.

  12. Ensure local communities retain agency over data centre development decisions, local government planning powers are not overridden, and AI developers bear the environmental, infrastructure and community costs of their business.

    Data centres provide few opportunities for ongoing employment while imposing energy, water and infrastructure demands on host communities. Governments should ensure local communities retain agency over development decisions and local government planning powers are not overridden, and that AI developers bear the environmental, infrastructure and community costs of their business.

  13. Ensure AI benefits are shared equitably with Aboriginal and Torres Strait Islander peoples through community-led approaches.

    AI benefits must be shared equitably with Aboriginal and Torres Strait Islander peoples. Limited access to digital infrastructure and AI literacy in remote communities could create further disadvantage for Aboriginal and Torres Strait Islander peoples and undermine Closing the Gap strategies by widening disparities in employment, skills development and economic participation. There must be community-led approaches to ensure Aboriginal and Torres Strait Islander peoples also benefit from AI.

  14. Establish a government-coordinated sovereign AI initiative with unions, universities, industry experts, superannuation and Australian technology industries, developing foundation and industry-specific models for Australian cultural, social and economic needs while using and protecting Australian data.

    Australian unions are calling for a government-coordinated sovereign AI initiative bringing together unions, universities, industry experts, the superannuation sector and the Australian tech industry to develop Australian AI capability. This would include developing Australian AI foundation models and industry-specific models designed for Australian cultural, social and economic needs while using and protecting Australian data.

  15. Prioritise government AI systems that are explainable, auditable and subject to strong governance, including declared version evolution.

    Government should prioritise AI Public Services International, Digitalisation: A Union Action Guide for Public Services, Work and Workers, 7 January 2025, https://publicservices.international/resources/digital-publication/digitalisation-br-a-union-action-guide-for-public-services-work-and-workers?id=11767&lang=en#2-work-workers-rights-and-workplace-governance systems that are explainable, auditable and subject to strong governance controls, including declared version evolution.

  16. Have public sector AI procurement support Australian sovereign capability, including domestic expertise, infrastructure, data storage, governance capability and public sector technical capacity.

    Public sector AI procurement should support Australian sovereign capability. Sovereign capability should include domestic expertise, infrastructure, data storage, governance capability and public sector technical capacity.

  17. Make a strong APS commitment to AI technical skills development, workforce capability and retention.

    The APS should make a strong commitment to AI technical skills development, workforce capability and retention to maintain critical public sector skills and expertise.

  18. Where practicable, have public sector AI systems provided by Australian public or private entities to reduce reliance on overseas entities.

    Where practicable, public sector AI systems should be provided by Australian public or private entities to reduce reliance on overseas entities and strengthen national capability.

  19. Test investment in AI against the opportunity cost of other national priorities, such as electrification and renewables.

    Investment in AI should be tested against the opportunity cost of other national priorities that may deliver greater returns, such as electrification and renewables priorities.

  20. Subject Australian data used in delivering public services to strong sovereignty, stewardship and governance requirements.

    Australian data used in the delivery public services should be subject to strong sovereignty, stewardship and governance requirements.

  21. Host and process AI systems trained on Australian citizens’ data for Australian citizens’ benefit onshore in Australian data centres, safeguarding against offshore access and secondary use.

    AI systems trained on the data of Australian citizens for the benefit of Australian citizens must be hosted and processed onshore in Australian data centres, with safeguards against offshore access and secondary use.

  22. Give individuals meaningful control over use of their personal information, including protection from its use to train AI models without informed consent.

    Individuals should have meaningful control over how their personal information is used, including protection from its use to train AI models without informed consent.

  23. Give Aboriginal and Torres Strait Islander peoples meaningful control over the use, management and sharing of their cultural and intellectual data.

    Aboriginal and Torres Strait Islander peoples should have meaningful control over the use, management and sharing of their cultural and intellectual data.

  24. Give workers consultation rights through Enterprise Agreements and the Fair Work Act, amend the Act for statutory pre-decision AI consultation, and consult workers and representatives early and continuously with genuine influence.

    Before decisions are made on AI-related workplace change, workers must have a right to consultation through Enterprise Agreements and the Fair Work Act 2009. This includes amending the Fair Work Act 2009 to provide a statutory right to pre-decision consultation on AI. Consultation with workers and their representatives should occur at the earliest possible stage and be ongoing, with a genuine opportunity for workers to influence outcomes.

  25. Provide universal APS AI literacy and role-specific training, ongoing retraining and upskilling, and broader skills to preserve policy advice, program management and research capability against AI overreliance.

    Universal AI literacy training should be provided to all APS employees, supported by role- specific training as required. There should be opportunities for ongoing retraining and upskilling as new opportunities and roles emerge. This should go beyond AI-specific skills and ensure that core APS functions such as policy advice, program management and research are not diminished by overreliance on AI.

  26. Mitigate the impact of automation and job displacement in the APS and ensure workforce planning anticipates the effect of technological change.

    Government should actively seek to mitigate the impact of automation and job displacement in the APS and ensure workforce planning anticipates the effect of technological change.

  27. Guarantee retraining and upskilling in APS service-wide bargaining for employees whose roles are likely to be affected by AI, with affected workers able to elect retraining as new roles emerge.

    The CPSU is seeking a commitment from Government in APS service-wide bargaining to guarantee retraining and upskilling opportunities for employees in roles that are likely to be affected by AI. Affected workers can elect to be retrained as new roles emerge.

  28. Sustain service-wide investment in APS AI, data and digital expertise.

    Ongoing investment in cultivating service-wide expertise in AI, data and digital capabilities is essential to building and maintaining APS capability.

  29. Protect entry-level APS roles, guarantee meaningful work for new employees and preserve pathways for skills development.

    The APS must protect entry-level roles, guarantee meaningful work for new employees and preserve pathways for skills development.

  30. Preserve workers’ opportunities to exercise professional judgement, job control and autonomy.

    Workers must retain opportunities to exercise professional judgement and maintain job control and autonomy over their work.

  31. Do not hold employees responsible for errors they lacked the training, authority or visibility to prevent; place accountability for public-service AI with those deploying and governing it.

    APS employees should not be held responsible for errors that occur where they lacked sufficient training, authority or visibility to detect and prevent those errors. Accountability for AI used in public service delivery rests with the decision-makers deploying and governing those systems.

  32. Provide protections against work intensification as routine tasks are automated and remaining work becomes more complex.

    Protections are needed to prevent work intensification as routine tasks are automated, and remaining work becomes more complex.

  33. Protect workers against cognitive impacts of continued AI use, including lost critical-thinking skills and cognitive atrophy from overreliance.

    Protections are also needed to address potential cognitive impacts of continued AI use on workers, including loss of critical thinking skills and cognitive atrophy as a result of overreliance on AI.

  34. Ensure WHS legislation and workers' compensation schemes are fit for purpose in protecting workers from the physical, psychological and cognitive risks of AI at work.

    This includes ensuring WHS legislation and workers’ compensation schemes are fit for purpose in protecting workers from the physical, psychological and cognitive risks associated with the use of AI at work.

  35. Provide stronger central oversight and enforcement powers for AI in the APS, supported by independent auditing, to ensure consistent standards across agencies.

    Stronger central oversight and enforcement powers are needed, supported by independent auditing, to ensure consistent standards across the APS.

  36. Require early pre-decision APS AI consultation giving workers and unions genuine influence, including detailed and ongoing workforce discussions.

    Strong pre-decision consultation obligations must apply for the adoption of AI in the APS, with an opportunity for workers and their union to engage at the earliest possible stage with a genuine opportunity to influence the approach taken. Engagement should include opportunities for detailed and ongoing discussions with the workforce.

  37. Enforce public disclosure of APS AI use cases, impact-assessment summaries and governance, and clearly explain to workers and the public how AI affects services, workplace processes and decisions.

    APS agencies should be subject to enforceable transparency obligations including public disclosure of AI use cases, impact assessment summaries, and governance arrangements. Clear information should be provided to workers and the public about how AI systems influence service delivery, workplace processes and decisions.

  38. Strengthen independent monitoring, reporting and auditing of high-risk systems, with clear pathways for workers and the public to report concerns, errors and harms.

    Independent monitoring, reporting and auditing of high-risk systems should be strengthened and there should be clear pathways for workers and the public to report concerns, errors, harms and unintended consequences arising from AI use cases.

  39. Keep human decision-makers responsible and accountable for government decisions, with informed human oversight embedded in all decision-making processes.

    Human decision-makers must remain responsible and accountable for government decisions. Where AI is used to support decision-making processes, there is a risk of automation bias, whereby decision-makers may defer uncritically to AI-generated outputs. To mitigate this risk, informed human oversight must be embedded in all government decision-making processes.

  40. Ensure public servants retain the capability and authority to scrutinise and override AI-assisted outputs where necessary.

    Public servants must retain the capability and authority to scrutinise and override AI-assisted outputs where necessary, ensuring decisions reflect human judgement, accountability and adherence to legal and ethical obligations.

  41. Give individuals affected by AI-assisted decisions accessible, timely and enforceable rights to review and appeal.

    Individuals impacted by AI-assisted decisions must have accessible, timely and enforceable rights to review and appeal decisions.

  42. Inform individuals when AI has been used in decision-making and give them the option to opt out of AI use.

    Individuals should be informed when AI has been used in decision-making processes and have the option to opt-out of AI use.

  43. Require agencies to monitor reviews and complaints to identify and resolve failures in AI-assisted decision-making.

    Agencies must be required to monitor reviews and complaints to identify and resolve failures in AI- assisted decision-making processes.

  44. Keep APS record-keeping requirements in step with more complex decision-making so individuals can access the information needed to seek a review.

    APS record-keeping requirements must keep pace as decision-making processes become more complex, to ensure individuals have access to the information required to seek a review.

  45. Prioritise workforce capability so AI use enhances public services: train all APS employees in ethical and safe AI use, with role-specific training for high-risk or specialised applications.

    Government must prioritise building workforce capability to ensure the use of AI enhances the quality of public services and benefits the communities who rely on them. All APS employees must be trained in the ethical and safe use of AI in the workplace, with role- specific training for employees working with high-risk or specialised applications.

  46. Do not deploy AI systems unless employees have the skills, knowledge and support to use them safely and responsibly.

    AI systems should not be deployed unless employees have the skills, knowledge and support to use them safely and responsibly.

  47. Ensure AI complements rather than replaces the expertise, judgment and accountability of public servants.

    AI must complement, rather than replace, the expertise, judgment and accountability of public servants.

  48. Prioritise in-house public sector capability and invest in development or adoption of locally developed AI solutions wherever possible.

    Government should prioritise in-house capability and invest in the development or adoption of locally-developed AI solutions wherever possible.

  49. Require public-sector procurement transparency, auditability and access to information about vendor-developed service-delivery systems, with safeguards extending to outsourced work.

    Public sector procurement frameworks should require transparency, auditability, and access to information about vendor-developed systems used in government service delivery and safeguards must extend to outsourced work.

  50. Expand the AI Safety Institute from technical advice to government and give it appropriate resources and powers to oversee AI technology used in Australia.

    we continue to call for the AISI to be expanded from its role of providing technical advice to government and be given appropriate resources and powers to oversee AI technology being used in Australia.

  51. Investigate a licensing scheme for AI frontier models, managed by the AI Safety Institute, requiring AI developers and deployers to be licensed to deploy products into Australia.

    Australian unions also recommend investigating a licensing scheme for AI frontier models, requiring AI developers and deployers to be licensed to deploy their products into Australia, to be managed by the AISI.

  52. Give the AI Safety Institute enforcement powers, with commensurate resources, to coordinate oversight, close regulatory gaps and address harms between existing regulatory boundaries.

    The AI Safety Institute should be given enforcement powers to coordinate oversight, close regulatory gaps and address harms that fall between existing regulatory boundaries. This must be met with commensurate resources.

  53. Provide regulators with adequate funding, resourcing and support to build capability and technical expertise.

    Regulators vary in powers and resourcing and require adequate funding, resourcing and support to build the capability and technical expertise required.

  54. Adapt Australia's industrial relations frameworks to keep pace with AI adoption, including pre-decision consultation, transparency, privacy protections from workplace surveillance, WHS protections and protection from discrimination and bias.

    Australia’s industrial relations frameworks must be adapted to keep pace with the increasing adoption of AI in workplaces. This includes meaningful pre-decision consultation on AI, transparency on AI use, privacy protections including from workplace monitoring and surveillance, workplace health and safety protections and protections from discrimination and bias.

  55. Regulate data centres to minimise pressure on energy grids, including renewable energy requirements, demand-management measures and aligning high-processing activities with lower grid demand.

    ensure data centres minimise pressure on energy grids, including through renewable energy requirements, demand-management measures and alignment of high- processing activities with periods of lower grid demand.

  56. Regulate data centre emissions and require lower-emissions backup power, including battery storage and renewable alternatives to diesel generators.

    regulate emissions from data centres and require the adoption of lower-emissions backup power sources, including battery storage and other renewable energy alternatives to diesel generators.

  57. Regulate data-centre water use: require treated effluent, air or closed-loop cooling where feasible; limit town and freshwater use and promote rainwater harvesting and recycling.

    regulate water use by requiring the use of treated effluent, air cooling or closed-loop cooling systems where feasible; limiting reliance on town water supplies and freshwater sources and promoting rainwater harvesting and water recycling.

  58. Review and update noise pollution standards to address the low-frequency and continuous noise generated by data centres.

    review and update noise pollution standards to address the low frequency and continuous noise generated by data centres.

  59. Ensure the Australian Federal Police have sufficient powers and resources to investigate and prosecute AI-enabled harms such as deepfake abuse, scams and spam.

    The Australian Federal Police should have sufficient powers and resources to investigate and prosecute AI-enabled harms, such as deepfake abuse, scams and spam.

  60. Subject AI systems that generate hate speech, discriminatory or other harmful material to strict penalties and require developers to implement corrective measures.

    AI systems that generate hate speech, discriminatory content or other harmful material should be subject to strict penalties and require developers to implement corrective measures to prevent further harm.

  61. Give individuals a legally enforceable right to prevent their personal information being used to train or operate AI systems.

    Individuals should have a legally enforceable right to prevent their personal information from being used to train or operate AI systems.

  62. Give creators the right to withhold consent to AI training use of their cultural, artistic and intellectual property, with penalties for non-compliant developers and deployers.

    Creators should have the right to withhold consent for the use of their cultural, artistic and intellectual property to train AI models, with penalties for developers and deployers of AI who do not comply with these requests.

  63. Make legislative protection retrospective for data already used to train existing AI models, with compensation and removal rights for affected creators.

    Legislative protection should be retrospective for data that has already been used to train existing AI models, with compensation and removal rights for affected creators.

  64. Do not let AI-generated outputs create new, independent IP ownership for the user where derived from existing copyrighted training data.

    AI-generated outputs should not create new, independent IP ownership for the user where it has been derived from existing copyrighted training data.

  65. Provide strong oversight, regular independent auditing, transparency and ongoing monitoring of AI in security vetting.

    CPSU members have similarly raised concerns that security clearance processes may create barriers for some Australians from culturally diverse backgrounds.18 There is a real risk that AI systems reinforce or even amplify existing biases, undermining APS workforce capability and damaging our national interest. Strong oversight, regular independent auditing, transparency in decision-making processes, and ongoing monitoring for unintended impacts is needed.

  66. Consider collecting and publishing deidentified demographic data on security clearance applications and outcomes to identify whether specific cohorts are disproportionately affected.

    Consideration should also be given to collecting and publishing deidentified demographic data on security clearance applications and outcomes to identify whether specific cohorts are unintentionally being disproportionately affected.

Read the original

Open in the interactive guide