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 352

Ms Litsa Roberts

Submitter group: Individuals.

Position labels are not shown for individual submitters.

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.

Proposes five recommendations: a national AI operating model and architecture, a mandatory strategic control assessment for critical AI, rights and economy infrastructure for AI and copyright, a statutory governability and accountability duty for high-impact AI, and enforceable public value requirements for major AI infrastructure investment.

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. Recommends a statutory governability and accountability duty for consequential, high-impact AI.

    Australia should establish a statutory AI Governability and Accountability Duty for consequential and high-impact AI, with obligations proportionate to the system’s capability, autonomy, access, scale and consequence.

  2. Recommends the Office of AI publish and maintain a National AI Operating Model and Architecture as a practical national coordination mechanism.

    The Australian Government should direct the Office of AI [3] to publish a National AI Operating Model and Architecture within 12 months and maintain it as an evolving national governance instrument. It should provide a practical mechanism for coordinating national AI capability, governance, accountability and resilience, not another statement of principles.

  3. Argues that data residency alone is not sovereignty, because decision rights, IP, model control and economic value may remain elsewhere.

    Data residency alone is not sovereignty. A system may be hosted in Australia while material decision rights, intellectual property, model control, operational knowledge and economic value remain elsewhere.

  4. Argues that rights and permissions should be identified and resolved before use where reasonably possible, rather than disputed afterwards.

    The practical principle is rights before scale: where rights can reasonably be identified and resolved before use, they should not routinely be left to retrospective dispute after potentially millions of uses or transactions.

  5. Recommends that affected people receive notice, meaningful explanations, information correction, challenge and genuine reconsideration by a competent human authorised to change the result.

    Affected people should be able to know when AI materially influenced the decision, understand the basis of the outcome to a meaningful degree, correct relevant information, challenge the decision and obtain genuine reconsideration by a competent human with sufficient information, discretion and authority to change the result.

  6. Recommends that major AI infrastructure receiving significant public support be required to prepare a Public Value Statement.

    Where major AI infrastructure receives material Commonwealth, state or territory funding, incentives, procurement commitments, approvals, regulatory facilitation or other significant public support, proponents should be required to prepare a Public Value Statement with measurable commitments proportionate to the scale, impact and support associated with the investment.

Recommendations

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

  1. Set a public-interest mandate for AI to improve human and national outcomes while protecting rights, privacy, safety, dignity and meaningful choice; embed human rights in design and governance.

    Government should set a clear public-interest mandate: AI should improve human and national outcomes while protecting rights, privacy, safety, dignity and meaningful human choice. Human rights should be design and governance requirements, not retrospective considerations after deployment.

  2. Calibrate controls to function, capability, autonomy, accessible resources, consequences and reversibility, affected people’s vulnerability, service criticality, and the ability to reconstruct, explain, challenge, intervene, stop and remedy outcomes.

    At minimum, the applicable control level should consider: • the system’s function and capability, including whether it predicts, generates, recommends, plans, acts or controls physical systems; • the degree of autonomy and the decisions, data, tools, financial resources, production systems or physical environments it can access; • the severity, scale and reversibility of potential consequences, including misuse, cyber, fraud, deception, privacy, discrimination, physical harm; • the vulnerability of people affected and the criticality of the service or infrastructure; and • the ability to identify, reconstruct, explain, challenge, intervene, stop and remediate outcomes.

  3. Require nationally consistent legislation for differentiated AI governance with obligations scaling to capability, autonomy, access and consequence.

    Nationally consistent legislation should require differentiated AI governance, with obligations scaling to capability, autonomy, access and consequence.

  4. Have regulators publish and periodically update practical classification and assurance thresholds so lower-risk AI is not burdened by controls designed for high-impact systems.

    Regulators should publish and periodically update practical classification and assurance thresholds so lower-risk AI is not burdened by controls designed for high-impact systems.

  5. Direct the Office of AI to publish a National AI Operating Model and Architecture within 12 months and maintain it as an evolving national governance instrument.

    The Australian Government should direct the Office of AI [3] to publish a National AI Operating Model and Architecture within 12 months and maintain it as an evolving national governance instrument.

  6. Have the national operating model identify needed capabilities, named accountable policy/assurance/regulatory/incident/resilience owners, decision rights, escalation thresholds and mandatory controls for critical and high-consequence AI.

    It should identify, at a minimum:  the national AI capabilities Australia needs to plan, build, retain or reliably access;  the named accountable owner for each major policy, assurance, regulatory, incident-response and resilience function;  the decision rights, escalation thresholds and mandatory controls applying to critical and high-consequence AI;

  7. Map dependencies across computing, cloud, data, models, algorithms, agents, APIs, identity/access, connectivity, cybersecurity, energy/water, supply chains and critical infrastructure; identify lifecycle standards, assurance, validation and regulatory controls.

    dependencies across compute and cloud, data, models, algorithms, agents, APIs and integration interfaces, identity and access, connectivity, cyber security, energy, water, supply chains and critical infrastructure;  where technical, rights and interoperability standards, assurance, validation, regulatory controls apply across the AI lifecycle;

  8. Specify government/jurisdiction coordination for systemic, emerging and cumulative risks; establish monitoring, assurance, decision, escalation, remediation, reporting and review rhythms, with published capability, resilience, safety, rights, contestability, incident and public-value measures.

    cross-government and cross-jurisdiction coordination pathways for systemic, emerging, cumulative and compounding risks and dependencies;  the national operating rhythm for monitoring, assurance, decision-making, incident escalation, remediation, reporting and periodic review; and  published measures for capability, resilience, safety, human safeguards, rights and economy infrastructure, contestability, incident response, remediation and public value.

  9. Define requirements, standards, interfaces, evidence and accountability without prescribing one platform or rebuilding existing capabilities; admit interoperable, assured, rights-protecting, secure and accountable systems while avoiding vendor or proprietary-standard dependence.

    The National AI Operating Model and Architecture should not prescribe a single technology platform or require government to rebuild capabilities that already exist. It should define national requirements, standards, interfaces, evidence and accountability, allowing government, commercial, open and sector-specific systems to participate where they meet interoperability, assurance, rights-protection, security and accountability requirements. The objective should be a federated and interoperable ecosystem that avoids unnecessary dependence on any single vendor or proprietary standard.

  10. Have the Office of AI maintain a public AI Governance, Capability and Rights Dashboard covering national measures, dependencies, control gaps, rights and economy infrastructure development and remediation.

    The Office of AI should maintain a public AI Governance, Capability and Rights Dashboard showing material national measures, dependencies, control gaps, rights and economy infrastructure development and remediation activity.

  11. Build the public dashboard on the DTA transparency register; show owners, target dates, implementation, dependencies, weaknesses, overdue actions and historical capability, resilience, safety, human-safeguard, rights and public-value trends.

    This should build on, rather than duplicate, the DTA’s existing central register of Commonwealth AI transparency statements. [6] The dashboard should show accountable owners and target dates, implementation status, material dependencies and control weaknesses, overdue actions, and historical trends across capability, resilience, safety, human safeguards, rights and public value.

  12. Update the public dashboard at least quarterly, reflect material changes promptly where practicable, and add incident/emerging-risk updates unless disclosure creates security, privacy, safety or legitimate commercial-sensitivity risks.

    It should be updated at least quarterly, with material status changes reflected promptly where practicable. Significant incidents or emerging risks should trigger additional updates where public disclosure would not create a national security, privacy, safety or legitimate commercial-sensitivity risk.

  13. Have the responsible Minister publish an annual implementation statement identifying material capability gaps, unresolved dependencies and significant control weaknesses, with accountable owners, remediation actions, timeframes and escalation status.

    The responsible Minister should publish an annual implementation statement identifying material capability gaps, unresolved dependencies and significant control weaknesses, together with accountable owners, remediation actions, target timeframes and escalation status.

  14. Escalate materially overdue implementation actions rather than carrying them forward.

    Materially overdue actions should be escalated rather than simply carried forward.

  15. Require a mandatory AI Strategic Control Assessment for Commonwealth procurement and designated nationally critical or high-consequence AI uses, before approval, material expansion or renewal.

    For Commonwealth procurement and designated nationally critical or high-consequence uses of AI, including systems supporting critical infrastructure, essential government functions or significant rights and public interests, I recommend a mandatory AI Strategic Control Assessment before approval, material expansion or renewal.

  16. Base critical-AI designation on consequence, dependency, concentration, substitutability, continuity risk and ability to exit/adapt, rather than supplier nationality, company size or model class alone.

    Designation should be based on consequence, dependency, concentration, substitutability, continuity risk and the capacity to exit or adapt — not on supplier nationality, company size or model class alone.

  17. Assess independent understanding/assurance of systems and dependencies, Australian control of sensitive data, permissions and access, and essential-service continuity if suppliers, models, services or geopolitical conditions change.

    The assessment should answer six questions: 1. Can the system and its material dependencies be independently understood and assured to the level required by the use case? 2. Can sensitive Australian data, permissions and access be controlled under Australian requirements? 3. Can essential operations continue if a supplier, model, service or geopolitical condition changes?

  18. Assess data/workload portability and exit without unreasonable lock-in, Australian skills to operate/assure/adapt/replace critical components, and retained Australian economic, research or knowledge value.

    4. Can data and workloads be moved, and can the organisation exit without unreasonable operational or technical lock-in? 5. Does Australia retain people with enough capability to operate, assure, adapt or replace critical components? 6. Does the arrangement retain meaningful Australian economic, research or knowledge value?

  19. Where appropriate, incorporate the strategic control test into critical-infrastructure, sector-regulatory, funding and national-security frameworks, aligning states and territories where responsibilities are shared.

    Beyond Commonwealth procurement, the same test should be incorporated where appropriate into critical-infrastructure, sector-regulatory, funding and national-security frameworks, with state and territory alignment where responsibilities are shared.

  20. For Commonwealth uses, embed the assessment in procurement and architecture assurance with risk-calibrated minimum controls, and require written senior-executive approval and independent review for material exceptions.

    For Commonwealth uses, the assessment should be embedded in procurement and architecture assurance, with minimum controls calibrated by risk. Material exceptions should require written senior-executive approval and independent review.

  21. Include proportionate portability, continuity, audit, change-notification and exit obligations in contracts.

    Contracts should include proportionate portability, continuity, audit, change-notification and exit obligations.

  22. For nationally critical private-sector uses, carry equivalent obligations through the relevant regulatory, critical-infrastructure or funding instrument rather than relying on voluntary adoption.

    For nationally critical private-sector uses, equivalent obligations should be carried through the relevant regulatory, critical-infrastructure or funding instrument rather than relying on voluntary adoption.

  23. Develop national interoperable rights and economy infrastructure supporting identification, verification, authorisation, transactions, attribution, compensation and audit of rights, permissions and legitimate value flows at digital scale.

    3. Rights and Economy Infrastructure: Develop national, interoperable Rights and Economy Infrastructure that strengthens Australian sovereignty by enabling rights, permissions and legitimate value flows to be identified, verified, authorised, transacted and audited at digital scale — across identity, attribution, provenance, entitlement, consent, authorisation, licensing, usage, compensation and audit.

  24. Preserve rights-holder choice; avoid single-platform dependence and proprietary lock-in; make rights and economy infrastructure available across creators, rights holders, societies, technology providers, AI developers, platforms, businesses and government.

    It should preserve rights-holder choice, avoid dependence on a single government or commercial platform, and minimise lock-in to one provider or proprietary standard. It should be an economy-wide national capability for creators, rights holders, collecting societies, technology providers, AI developers, platforms, businesses and public-sector users.

  25. Make generative-AI rights and permissions increasingly machine-readable and, where legally/technically appropriate, resolvable before authorised generation/use, with applicable usage, attribution, licensing, compensation and audit records.

    For generative AI, rights and permissions should increasingly be machine-readable and, where legally and technically appropriate, capable of being resolved at or before authorised generation or use, with relevant usage, attribution, licensing, compensation and audit records created where applicable.

  26. Have the Attorney-General's Department and Office of AI jointly convene a 12-month AI National Standards and Pilot Program with rights holders, collecting societies, technology companies, standards bodies, researchers and affected sectors.

    The Attorney-General’s Department and Office of AI should jointly convene a 12-month AI National Standards and Pilot Program with rights holders, collecting societies, technology companies, standards bodies, researchers and affected sectors.

  27. Run two or three pilots in copyright-intensive domains to test whether rights, permissions and associated economic information can be exchanged reliably across organisations and platforms.

    Two or three pilots in copyright-intensive domains should test whether rights, permissions and associated economic information can be exchanged and acted upon reliably across organisations and platforms.

  28. Evaluate rights pilots for accurate identity, provenance, entitlements and permissions; interoperability and digital authorisation/licensing/transactions; and reliable records of authorised usage, attribution, licensing and applicable compensation.

    They should evaluate:  accuracy of identity, provenance and entitlement information, and expression and verification of permissions and restrictions;  interoperability among rights holders, intermediaries, platforms and AI systems, including support for authorisation, licensing and legitimate transactions at digital scale;  reliable recording of authorised usage, attribution, licensing and compensation where applicable;

  29. Evaluate rights pilots for auditability, traceability, dispute resolution, corrections, affordability/accessibility for smaller creators and rights holders, and technical, platform or commercial lock-in risks.

    auditability and traceability, dispute resolution and correction mechanisms; and  accessibility and affordability for smaller creators and rights holders, and the risk of technical, platform or commercial lock-in.

  30. Publish rights and interoperability requirements, standards recommendations and pilot evidence without endorsing a single commercial implementation.

    The Government should publish rights and interoperability requirements, standards recommendations and pilot evidence without endorsing a single commercial implementation.

  31. Progressively require rights metadata, permissions, licensing, usage records and auditability for licensed Australian content in Commonwealth procurement once workable standards are demonstrated.

    Commonwealth procurement can accelerate adoption by progressively requiring appropriate rights metadata, permissions, licensing, usage records, and auditability for licensed Australian content once workable standards are demonstrated.

  32. If rights and interoperable mechanisms prove viable, determine where proportionate mandatory requirements are warranted, with the Commonwealth legislating within its jurisdiction and working with states and territories toward nationally consistent requirements.

    If rights and interoperable mechanisms prove technically and commercially viable, government should determine where proportionate mandatory requirements are warranted. The Commonwealth should legislate within its jurisdiction and work with states and territories toward nationally consistent requirements.

  33. Have existing regulators enforce record-keeping, audit and remediation obligations, with organisations remaining accountable where third-party rights-management or AI services are used.

    Existing regulators should enforce record-keeping, audit and remediation obligations, while organisations remain accountable where third-party rights-management or AI services are used.

  34. Establish a statutory AI Governability and Accountability Duty for consequential and high-impact AI, with obligations proportionate to capability, autonomy, access, scale and consequence.

    Australia should establish a statutory AI Governability and Accountability Duty for consequential and high-impact AI, with obligations proportionate to the system’s capability, autonomy, access, scale and consequence.

  35. Protect relevant rights and freedoms, including privacy, equality, procedural fairness, accessibility, dignity and effective remedy, in the statutory baseline for consequential AI, with heightened safeguards for children, vulnerable people or significant power imbalances.

    For consequential AI, the statutory baseline should protect relevant rights and freedoms including privacy, equality and non-discrimination, procedural fairness, accessibility, dignity and effective remedy, with heightened safeguards where children, vulnerable people or significant power imbalances are involved.

  36. Require a Governability Gate before consequential or high-impact AI deployment or material expansion, evidencing purpose, necessity, authority, validation, performance, data quality, privacy, cybersecurity, fairness, oversight, rights, traceability, monitoring, incident response and remedies.

    Before consequential or high-impact AI is deployed or materially expanded, it should pass a pre-deployment Governability Gate. The Gate should require proportionate evidence of purpose, necessity, legal authority, validation, performance, data quality, privacy, cyber security, fairness, human oversight, rights protection, traceability, monitoring, incident response, challenge and remedy.

  37. Make the pre-deployment approval outcome explicit: approve, approve with conditions, constrain or do not deploy.

    The approval outcome should be explicit: APPROVE → APPROVE WITH CONDITIONS → CONSTRAIN → DO NOT DEPLOY

  38. Reassess material model, algorithm, data, supplier, system-access, autonomy, decision-logic or use-case changes; maintain governance throughout the lifecycle rather than only at approval.

    Material changes to the model, algorithm, data, supplier, system access, autonomy, decision logic or use case should trigger reassessment. Governance should continue throughout the lifecycle rather than becoming a one-time approval exercise.

  39. Consider cumulative and compounding consequences in governability assessments, particularly where repeated or interconnected AI decisions amplify disadvantage or harm.

    Governability assessments should also consider cumulative and compounding consequences and impacts, particularly where repeated or interconnected AI decisions can amplify disadvantage, discrimination, exclusion, financial harm or loss of opportunity over time.

  40. Make factual integrity and authenticity part of the Governability Gate where AI can materially influence decisions, public information or democratic processes.

    Where AI can materially influence decisions, public information or democratic processes, factual integrity and authenticity should form part of the Governability Gate.

  41. Where AI materially influences decisions, public information or democracy, support technically appropriate provenance/authenticity, identify material uncertainty, and enable detection and correction of fabricated, manipulated or materially false information.

    Where AI can materially influence decisions, public information or democratic processes, factual integrity and authenticity should form part of the Governability Gate. Systems should support provenance and authenticity where technically appropriate, identify material uncertainty, and make it possible to detect and correct fabricated, manipulated or materially false information.

  42. Require dashboards for designated consequential or high-impact AI, reviewed by boards, governing bodies or accountable executives at least quarterly and more frequently where risk warrants.

    Organisations deploying designated consequential or high-impact AI should maintain an AI Governance Dashboard reviewed by the board, governing body or accountable executive at least quarterly, and more frequently where risk warrants.

  43. Dashboard reporting should cover realised/forecast value, cost, computing/vendor dependencies, autonomy/access, data/rights/regulation, incidents/exceptions/overrides/near misses, human interventions, stakeholder/operational impacts, executives, remediation and next decision gates.

    The dashboard should provide visibility over:  value realised versus forecast;  cost-to-serve, compute and material vendor dependency;  autonomy and system access;  data sensitivity, rights impacts and regulatory exposure;  incidents, exceptions, overrides and near misses;  human review and intervention rates;  customer, employee, community and operational impacts; and  the accountable executive, outstanding remediation and next decision gate.

  44. For material AI, require governing bodies explicitly to continue, continue conditionally, constrain, suspend or retire systems.

    For material AI, governing body should make an explicit decision: CONTINUE → CONTINUE WITH CONDITIONS → CONSTRAIN → SUSPEND → RETIRE

  45. Designate at minimum materially consequential AI in employment/recruitment, healthcare, credit/insurance/housing, education, benefits/essential services, law enforcement/biometrics/justice, elections, material algorithmic pricing, and critical infrastructure/autonomous physical systems.

    Designated high-consequence domains should include, at minimum, AI materially used in employment and recruitment; healthcare and medical decisions; credit, insurance and housing; education; government benefits and essential services; law enforcement, biometrics and justice; elections and democratic processes; material algorithmic pricing; and critical infrastructure or autonomous physical systems.

  46. Keep the designated high-consequence domain list reviewable as technology, capability and evidence evolve.

    The list should be reviewable as technology, capability and evidence evolve.

  47. Treat AI materially affecting recruitment, hiring, pay/remuneration, performance assessment, promotion, workforce allocation or termination as consequential.

    AI materially affecting recruitment, hiring, pay, remuneration, performance assessment, promotion, workforce allocation or termination should be treated as consequential.

  48. Require adverse-impact and proxy-discrimination testing for protected attributes including sex and age, and intersectional effects such as older women’s unequal pay or less favourable employment outcomes.

    Organisations should conduct adverse-impact and proxy-discrimination testing for protected attributes including sex and age, and for intersectional effects such as discrimination affecting older women, including unequal pay or less favourable employment outcomes.

  49. Test pay equity across salary recommendations, remuneration, bonuses, promotion and other material outcomes.

    Pay-equity testing should examine salary recommendations, remuneration, bonuses, promotion and other material outcomes.

  50. Give applicants and employees genuine human review where AI materially affects an adverse decision, with employers remaining accountable where third-party AI systems are used.

    Applicants and employees should have genuine human review where AI materially affects an adverse decision. Employers should remain accountable where third-party AI systems are used.

  51. Do not permit organisations to use historical workforce data, algorithmic scoring or AI-generated recommendations to perpetuate or justify discriminatory employment outcomes that would be unlawful if imposed by a human.

    An organisation should not be permitted to use historical workforce data, algorithmic scoring or AI-generated recommendations to perpetuate or justify discriminatory employment outcomes that would be unlawful if imposed directly by a human decision-maker.

  52. Strengthen dynamic or personalised pricing safeguards where essential goods or services, vulnerability, sensitive or inferred personal information, or material unfairness or anti-competitive risks are involved.

    Dynamic or personalised pricing should be subject to stronger safeguards where it affects essential goods or services, exploits vulnerability, uses sensitive or inferred personal information, or creates material risks of unfair or anti-competitive outcomes.

  53. Ensure consumers are not charged materially more solely because an AI system infers desperation, vulnerability or an inability to shop around.

    Consumers should not be charged materially more solely because an AI system infers desperation, vulnerability or an inability to shop around.

  54. Let regulators establish sector-specific thresholds, evidentiary tests and safe harbours.

    Regulators should be able to establish sector-specific thresholds, evidentiary tests and safe harbours.

  55. Treat AI materially influencing elections or democratic participation as high-consequence; address deceptive synthetic content, impersonation, false voting information, voter suppression, coordinated automated influence, provenance and disclosure.

    AI materially used to influence elections or democratic participation should be treated as high-consequence. Safeguards should address materially deceptive synthetic content, impersonation, false voting information, voter suppression, coordinated automated influence activity, provenance and appropriate disclosure.

  56. Make rapid correction, takedown and regulator intervention available where serious electoral harm is occurring, while protecting legitimate satire, parody, opinion and democratic expression.

    Rapid correction, takedown and regulator intervention should be available where serious electoral harm is occurring, while protecting legitimate satire, parody, opinion and democratic expression.

  57. Establish a statutory Right of Challenge (notice, explanation, correction, challenge, human review, remedy, audit) where AI materially influences a decision affecting a person's rights, livelihood, safety, essential services or opportunities.

    Where AI materially influences a decision affecting a person’s rights, livelihood, safety, essential services or opportunities, there should be a statutory Right of Challenge: NOTICE → EXPLANATION → CORRECTION → CHALLENGE → HUMAN REVIEW → REMEDY → AUDIT

  58. Enable affected people to know of material AI involvement, understand outcomes meaningfully, correct information, challenge decisions and obtain genuine reconsideration by a competent human with information, discretion and authority to change results.

    Affected people should be able to know when AI materially influenced the decision, understand the basis of the outcome to a meaningful degree, correct relevant information, challenge the decision and obtain genuine reconsideration by a competent human with sufficient information, discretion and authority to change the result.

  59. Retain sufficient decision records to reconstruct material AI-influenced decisions, including system version, inputs, output, decision rules, accountable decision-maker, whether the output was accepted or overridden, and the outcome.

    Sufficient decision records should be retained to reconstruct material decisions, including the relevant system or model version, material inputs, recommendation or output, applicable decision rules, accountable decision-maker or reviewer, whether the AI output was accepted or overridden, and the final outcome.

  60. Empower regulators to inspect evidence, require independent audits, mandate remediation, require reconsideration or compensation through appropriate legal mechanisms, restrict functionality, suspend deployment or prohibit use where serious deficiencies remain.

    Relevant regulators should have powers to inspect evidence, require independent audits, mandate remediation, require affected people to be reconsidered or compensated through the appropriate legal mechanism, restrict system functionality, suspend deployment or prohibit continued use where serious deficiencies remain.

  61. Require prompt notification to the relevant regulator of material incidents involving designated high-impact AI.

    Material incidents involving designated high-impact AI should trigger prompt notification to the relevant regulator rather than waiting for routine reporting cycles.

  62. Impose substantial civil penalties for serious, systemic or repeated breaches, proportionate to scale, benefit and harm and sufficient to deter rather than become a business cost.

    Serious, systemic or repeated breaches should attract substantial civil penalties proportionate to organisational scale, benefit obtained and harm caused. Penalties should be sufficient to deter misconduct rather than be absorbed as a cost of doing business.

  63. Apply applicable criminal liability, including lawful imprisonment, for knowingly continued deployment after identified serious unlawful/discriminatory impacts, falsified/concealed evidence, knowingly breached suspension orders, or AI facilitating existing crimes.

    Where an organisation knowingly continues to deploy a system after serious unlawful or discriminatory impacts have been identified, falsifies or conceals compliance evidence, knowingly breaches a regulator suspension order, or uses AI to facilitate conduct that is already criminal — including fraud, criminal cartel conduct, or relevant electoral or foreign-interference offences — responsible individuals should be subject to applicable criminal liability, including imprisonment where provided by law.

  64. Assess cumulative and compounding public-value effects across regions, infrastructure, ecosystems and communities, including whether successive investments build Australian value.

    Public-value assessment should look beyond a single project to cumulative and compounding consequences and impacts across regions, infrastructure, ecosystems and communities, and to whether successive investments create compounding Australian value.

  65. Require major AI infrastructure proponents receiving material Commonwealth/state/territory funding, incentives, procurement, approvals, facilitation or other significant public support to prepare Public Value Statements with measurable commitments proportionate to investment scale, impact and support.

    Where major AI infrastructure receives material Commonwealth, state or territory funding, incentives, procurement commitments, approvals, regulatory facilitation or other significant public support, proponents should be required to prepare a Public Value Statement with measurable commitments proportionate to the scale, impact and support associated with the investment.

  66. Where relevant, require public-value commitments on energy demand, new generation/storage/transmission, flexibility, water efficiency/source resilience, enabling-infrastructure costs, land use/planning, ecosystems/agriculture, cumulative regional impacts, local infrastructure and amenity.

    Those commitments should address, where relevant: • energy demand, additional generation, storage and transmission, demand flexibility, water efficiency and source resilience, and the allocation of enabling-infrastructure costs; • responsible land use, planning, ecosystems, agriculture, cumulative regional impacts, local infrastructure and amenity;

  67. Require relevant public-value commitments on security/resilience, supply chains, Australian jobs/skills, training/transition, research access, R&D, local suppliers, intellectual property and commercialisation.

    security and resilience, supply-chain dependencies, Australian jobs and skills, training and workforce transition, research access, Australian R&D, local suppliers, intellectual property and commercialisation;

  68. Require relevant commitments on competition, interoperability, portability, concentration risk, accessibility, regional participation, digital inclusion, community participation/benefit and transparent local reporting; assess successive investments’ enduring Australian capabilities and value.

    competition, interoperability, portability and concentration risk; accessibility, regional participation and digital inclusion; and meaningful community participation, measurable community benefit and transparent reporting against promised local outcomes. • Assessment should consider the extent to which successive investments build enduring Australian skills, research capability, intellectual property, infrastructure, supply chains, productivity and sovereign capability.

  69. Have each Public Value Statement identify a baseline, measurable target, accountable owner, timeframe and reporting frequency for every material commitment.

    Each Public Value Statement should identify a baseline, measurable target, accountable owner, timeframe and reporting frequency for every material commitment.

  70. Publish progress at defined intervals, subject to legitimate security, privacy and commercial-sensitivity limits.

    Progress should be public at defined intervals, subject to legitimate security, privacy or commercial-sensitivity limits.

  71. Have government report whether promised capability, resilience, environmental, infrastructure and community benefits were delivered for nationally significant projects.

    For nationally significant projects, government should report whether promised Australian capability, resilience, environmental, infrastructure and community benefits were actually delivered.

  72. Apply proportionate consequences for material/repeated non-performance: remediation, independent audit, reporting, adjustment/recovery of support, contractual remedies, future-approval conditions, regulator intervention and reconsideration of future public facilitation.

    Material or repeated non-performance should trigger proportionate consequences, including remediation, independent audit, additional reporting, adjustment or recovery of public support, contractual remedies, conditions on future approvals, regulator intervention and reconsideration of future public facilitation.

  73. Make public-value commitments enforceable where public money, regulatory facilitation or other material government support is provided.

    Public-value commitments should be enforceable where public money, regulatory facilitation or other material government support is provided, not non-binding statements of intent.

  74. Have the Office of AI publish an implementation map showing what can be done administratively and what requires legislation, sector rules, national standards, contractual controls or state and territory harmonisation.

    The Office of AI should publish an implementation map showing what can be done administratively and what requires Commonwealth legislation, sector rules, national standards, contractual controls or state and territory harmonisation.

  75. Implement the proposed operating-model, strategic-control and rights-pilot program within the first 12 months.

    A practical first 12-month program would be: 1. Publish the National AI Operating Model and Architecture, capability/dependency map, accountable owners and capability/consequence classification. 2. Introduce the AI Strategic Control Assessment for relevant Commonwealth procurement and designated nationally critical AI uses, with equivalent sector pathways where required. 3. Launch two or three national rights and economy infrastructure pilots and publish interoperability standards recommendations and evidence.

Read the original

Open in the interactive guide