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 477

Name withheld

Submitter group: Individuals.

Position labels are not shown for individual submitters.

An AI model identified the submitter group and chose the extracts. Counts do not measure importance or merit.

Open in the interactive guide

Extracts from the submission, in the submitter’s own words.

Extracts

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. it is noted that Australia relies on voluntary frameworks and existing laws, however these will be insufficient to protect the average Australian from potential harm.

  2. Existing Australian laws fail to protect the average person from AI harms because they were designed for analog-era risks and cannot address the novel structural vulnerabilities created by autonomous, opaque and adaptive AI systems.

  3. with the potential for deepfakes to collapse public trust in evidence, journalism, and democratic processes. Legally binding safeguards are not optional — they are foundational to maintaining a functioning democracy.

  4. Cloud Security Alliance, above n 13; Understanding AI, "Labs are struggling to keep frontier models under control" (13 August 2026) (noting many incidents occurred with cybersecurity guardrails deliberately disabled during evaluation).

  5. The Copyright Act 1968 (Cth) contains no specific exception for text and data mining (TDM) or AI model training, meaning training on Australian creative, cultural and media content must rely on existing fair dealing exceptions (which are narrow and do not clearly cover commercial AI development) or obtain licences from rights holders.

  6. AI middleware can fail silently, propagate errors across sectors, and cause high-impact harm with no clear regulatory remedy currently available.

Extracts from recommendations

Extracts chosen by an AI model, in the submitter’s own words. Listed in the same order as on its page in the interactive guide.

  1. The primary purpose of this submission is to urge the Joint Selection Committee to examine in detail the potential adverse impacts of artificial intelligence (AI) on ordinary Australians, and to consider how legislation and regulation can reduce these risks.

  2. In particular, the Committee is requested to use this opportunity to ensure that technology companies are required by law to be liable for and address societal and human impacts - not to focus primarily only company share value - when developing, deploying, and scaling AI systems.

  3. Future issues which must also be contemplated by the Joint Selection Committee include those where AI critical systems interact, including where one AI is trained on the outputs of another AI (potentially compounding errors).

  4. Amend the Copyright Act to establish a statutory licence for AI training on copyrighted material, administered through existing collecting societies (APRA AMCOS, PPCA, Copyright Agency) Require AI providers to register training datasets with a public registry (managed by IP Australia); creators can flag works for exclusion from training Mandate that AI providers publish summary lists of major training data sources (without disclosing proprietary models), subject to confidentiality protections where applicable (to avoid unwarranted reliance on broad confidentiality claims) Expected Outcome Ensures fair remuneration without blocking innovation; rates set by Copyright Tribunal

  5. Require AI providers to register training datasets with a public registry (managed by IP Australia); creators can flag works for exclusion from training

  6. Mandate that AI providers publish summary lists of major training data sources (without disclosing proprietary models), subject to confidentiality protections where applicable (to avoid unwarranted reliance on broad confidentiality claims)

  7. Shift evidentiary burden to AI developers to demonstrate lawful access when creators allege unauthorised training use

  8. Where infringement is established, courts must award account of profits derived from the unauthorised use (not merely damages)

  9. Provide 12-month grace period for models trained before commencement; require retrospective licensing or cessation

  10. Embed IPC principles in any AI training framework — separate consultation pathway for Traditional Owners, moratorium on culturally sensitive content

  11. The Joint Selection Committee is requested to prioritise implementation before early 2027 when the proposed Australian Standards for AI legislation commences, to avoid regulatory fragmentation and ensure Australia's cultural content is protected from extraction without recourse.

  12. AI-specific regulation should not be "last resort" but rather proportionate first response for high-risk applications, calibrated to Australia's innovation objectives while establishing baseline consumer and worker protections.

  13. The Joint Selection Committee should recommend a new statutory framework that establishes binding obligations rather than relying on voluntary standards.

  14. Establish four-tier risk system: (i) prohibited practices; (ii) high-risk systems; (iii) limited-risk transparency; (iv) minimal-risk voluntary.

  15. Apply to foreign providers whose AI systems affect Australian individuals or markets

  16. Machine-readable metadata embedded at point of generation of AI outputs, especially images and videos

  17. Apply to foreign AI providers whose systems affect Australian critical infrastructure, regardless of hosting location

  18. Civil penalties up to 6% of global turnover or A$50 million (whichever is higher); enforceable undertakings; injunction powers

  19. Enforcement should be allocated to existing regulators according to sectoral expertise, with cross-agency coordination supported through a central policy function within government.

  20. The Committee should ensure any new AI Act explicitly defines AI-mediated critical infrastructure to include: (i) Cross-domain data fusion systems operating across water, energy, health, or financial networks (ii) Agentic AI embedded in emergency call interpretation, dispatch, or triage workflows (iii) Cloud-based AI inference providers serving Australian critical infrastructure

  21. The Committee should also require that any acquisition of AI startups by infrastructure vendors undergo regulatory safety review prior to operational integration.

  22. The new legislation should prohibit the following AI applications without exception: Prohibition Social scoring Real-time remote biometrics Predictive policing Subliminal manipulation Emotional inference for employment/insurance Description AI systems evaluating individuals based on behaviour, socioeconomic status, or personal characteristics Live facial recognition in public spaces except by law enforcement, except for imminent terrorist threat or missing child investigation Risk assessment profiling individuals based on protected characteristics or behavioural patterns without human oversight AI designed to bypass conscious awareness to alter behaviour or decisions Using AI to infer emotions, mental state for employment screening or insurance underwriting

  23. Live facial recognition in public spaces except by law enforcement, except for imminent terrorist threat or missing child investigation Risk assessment profiling individuals based on protected characteristics or behavioural patterns without human oversight AI designed to bypass conscious awareness to alter behaviour or decisions Using AI to infer emotions, mental state for employment screening or insurance underwriting Comparison EU AI Act prohibition; aligns with human rights principles Balances security needs with privacy rights; requires judicial authorisation

  24. AI systems in the following sectors should be classified as high-risk and subject to mandatory compliance: Sector Specific Applications Required Safeguards Employment Recruitment screening, performance evaluation, termination decisions Loan approvals, credit scoring, insurance pricing Diagnostic tools, treatment recommendations, triage systems Human review; bias audit; appeal mechanism; data minimisation Credit & Finance Healthcare 8 Explainability; alternative dispute resolution; regular model validation Clinical validation; practitioner oversight; postmarket surveillance

  25. AI systems in the following sectors should be classified as high-risk and subject to mandatory compliance: Sector Specific Applications Required Safeguards Employment Recruitment screening, performance evaluation, termination decisions Loan approvals, credit scoring, insurance pricing Diagnostic tools, treatment recommendations, triage systems Human review; bias audit; appeal mechanism; data minimisation Credit & Finance Healthcare 8 Explainability; alternative dispute resolution; regular model validation Clinical validation; practitioner oversight; postmarket surveillance

  26. AI systems in the following sectors should be classified as high-risk and subject to mandatory compliance: Sector Specific Applications Required Safeguards Employment Recruitment screening, performance evaluation, termination decisions Loan approvals, credit scoring, insurance pricing Diagnostic tools, treatment recommendations, triage systems Human review; bias audit; appeal mechanism; data minimisation Credit & Finance Healthcare 8 Explainability; alternative dispute resolution; regular model validation Clinical validation; practitioner oversight; postmarket surveillance

  27. Justice System Critical Infrastructure Government Services AI Middleware & Data Fusion Sentencing guidance, parole risk assessment, bail determinations Energy grid control, transport management, emergency services coordination Welfare eligibility, visa decisions, tax assessment Cross-domain operational AI (water, energy, health, finance data integration) Judicial discretion preserved; transparency; independent audit Containment verification; kill-switch capability; incident reporting Right to human review; procedural fairness; reasons for decisions Single accountable entity designated; audit trail across vendor boundaries; mandatory testing before deployment; right to human override

  28. Justice System Critical Infrastructure Government Services AI Middleware & Data Fusion Sentencing guidance, parole risk assessment, bail determinations Energy grid control, transport management, emergency services coordination Welfare eligibility, visa decisions, tax assessment Cross-domain operational AI (water, energy, health, finance data integration) Judicial discretion preserved; transparency; independent audit Containment verification; kill-switch capability; incident reporting Right to human review; procedural fairness; reasons for decisions Single accountable entity designated; audit trail across vendor boundaries; mandatory testing before deployment; right to human override

  29. Justice System Critical Infrastructure Government Services AI Middleware & Data Fusion Sentencing guidance, parole risk assessment, bail determinations Energy grid control, transport management, emergency services coordination Welfare eligibility, visa decisions, tax assessment Cross-domain operational AI (water, energy, health, finance data integration) Judicial discretion preserved; transparency; independent audit Containment verification; kill-switch capability; incident reporting Right to human review; procedural fairness; reasons for decisions Single accountable entity designated; audit trail across vendor boundaries; mandatory testing before deployment; right to human override

  30. Justice System Critical Infrastructure Government Services AI Middleware & Data Fusion Sentencing guidance, parole risk assessment, bail determinations Energy grid control, transport management, emergency services coordination Welfare eligibility, visa decisions, tax assessment Cross-domain operational AI (water, energy, health, finance data integration) Judicial discretion preserved; transparency; independent audit Containment verification; kill-switch capability; incident reporting Right to human review; procedural fairness; reasons for decisions Single accountable entity designated; audit trail across vendor boundaries; mandatory testing before deployment; right to human override

  31. Compliance Requirements for High-Risk Systems: 1. Pre-deployment conformity assessment – Third-party evaluation before market entry 2. Risk management system – Documented processes for hazard identification and mitigation 3. Data governance – Training data quality, provenance documentation, bias testing 4. Technical documentation – Complete model architecture, decision logic, limitations 5. Logging and traceability – Audit trail for all automated decisions 6. Human oversight – Meaningful human involvement in high-stakes decisions 7. Accuracy and robustness – Minimum performance thresholds; adversarial testing 8. Cybersecurity – Protection against prompt injection, adversarial attacks, data poisoning

  32. Compliance Requirements for High-Risk Systems: 1. Pre-deployment conformity assessment – Third-party evaluation before market entry 2. Risk management system – Documented processes for hazard identification and mitigation 3. Data governance – Training data quality, provenance documentation, bias testing 4. Technical documentation – Complete model architecture, decision logic, limitations 5. Logging and traceability – Audit trail for all automated decisions 6. Human oversight – Meaningful human involvement in high-stakes decisions 7. Accuracy and robustness – Minimum performance thresholds; adversarial testing 8. Cybersecurity – Protection against prompt injection, adversarial attacks, data poisoning

  33. The legislation should mandate: Requirement Incident reporting Containment verification Kill-switch capability Middleware Accountability Mapping Frontier model evaluation Specification Mandatory notification within 24-72 hours of AI cybersecurity breach to AI regulator Independent certification that test environments are airgapped from production systems Verifiable emergency stop functionality for high-risk AI deployments AI providers must document all vendors in AI supply chain; designate single liable entity for end-to-end safety Pre-deployment safety assessment for models exceeding defined capability thresholds Testing window minimum Minimum 4-week pre-deployment safety testing period for frontier models prior to release Timeline Immediate upon enactment Quarterly audits Within 12 months Immediate upon enactment Before release

  34. The legislation should mandate: Requirement Incident reporting Containment verification Kill-switch capability Middleware Accountability Mapping Frontier model evaluation Specification Mandatory notification within 24-72 hours of AI cybersecurity breach to AI regulator Independent certification that test environments are airgapped from production systems Verifiable emergency stop functionality for high-risk AI deployments AI providers must document all vendors in AI supply chain; designate single liable entity for end-to-end safety Pre-deployment safety assessment for models exceeding defined capability thresholds Testing window minimum Minimum 4-week pre-deployment safety testing period for frontier models prior to release Timeline Immediate upon enactment Quarterly audits Within 12 months Immediate upon enactment Before release

  35. The legislation should mandate: Requirement Incident reporting Containment verification Kill-switch capability Middleware Accountability Mapping Frontier model evaluation Specification Mandatory notification within 24-72 hours of AI cybersecurity breach to AI regulator Independent certification that test environments are airgapped from production systems Verifiable emergency stop functionality for high-risk AI deployments AI providers must document all vendors in AI supply chain; designate single liable entity for end-to-end safety Pre-deployment safety assessment for models exceeding defined capability thresholds Testing window minimum Minimum 4-week pre-deployment safety testing period for frontier models prior to release Timeline Immediate upon enactment Quarterly audits Within 12 months Immediate upon enactment Before release

  36. The legislation should mandate: Requirement Incident reporting Containment verification Kill-switch capability Middleware Accountability Mapping Frontier model evaluation Specification Mandatory notification within 24-72 hours of AI cybersecurity breach to AI regulator Independent certification that test environments are airgapped from production systems Verifiable emergency stop functionality for high-risk AI deployments AI providers must document all vendors in AI supply chain; designate single liable entity for end-to-end safety Pre-deployment safety assessment for models exceeding defined capability thresholds Testing window minimum Minimum 4-week pre-deployment safety testing period for frontier models prior to release Timeline Immediate upon enactment Quarterly audits Within 12 months Immediate upon enactment Before release

  37. Pre-deployment safety assessment for models exceeding defined capability thresholds

  38. Minimum 4-week pre-deployment safety testing period for frontier models prior to release

  39. Users must be informed when interacting with AI (chatbots, voice assistants) to ensure visible disclosure on consumer-facing platforms

  40. Machine-readable watermarking + visible labels for AI-generated content

  41. 4.5 Transparency & Consumer Protection Obligation AI interaction disclosure Deepfake labelling Training data transparency Algorithmic impact assessment Right to explanation Appeal pathways Detail Users must be informed when interacting with AI (chatbots, voice assistants) to ensure visible disclosure on consumer-facing platforms Machine-readable watermarking + visible labels for AI-generated content Publishers of summary lists of major training data sources Public summary of potential biases and limitations Individuals affected by automated decisions entitled to meaningful explanation Formal complaints mechanism with independent review

  42. Individuals affected by automated decisions entitled to meaningful explanation

  43. 4.5 Transparency & Consumer Protection Obligation AI interaction disclosure Deepfake labelling Training data transparency Algorithmic impact assessment Right to explanation Appeal pathways Detail Users must be informed when interacting with AI (chatbots, voice assistants) to ensure visible disclosure on consumer-facing platforms Machine-readable watermarking + visible labels for AI-generated content Publishers of summary lists of major training data sources Public summary of potential biases and limitations Individuals affected by automated decisions entitled to meaningful explanation Formal complaints mechanism with independent review

  44. Specific right of action for AI-related harms (algorithmic discrimination, wrongful automated decisions)

  45. AI deployers must demonstrate system compliance when harm alleged

  46. Class action provisions for widespread AI harms

  47. Industry levy to compensate victims where liable party cannot pay

  48. Mechanism Statutory cause of action Reverse burden of proof Account of profits Collective redress Compensation fund Legal aid funding Implementation Specific right of action for AI-related harms (algorithmic discrimination, wrongful automated decisions) AI deployers must demonstrate system compliance when harm alleged Where infringement established, courts award profits derived from unauthorised use Class action provisions for widespread AI harms Industry levy to compensate victims where liable party cannot pay Priority listing for AI discrimination/harassment cases

  49. Lower compliance requirements for small-scale/non-commercial deployments (Small companies or tiny AI deployments shouldn’t have to meet the same compliance requirements as big banks, hospitals, or frontier-model developers)

  50. Temporary exemptions for testing innovative AI in controlled settings (Startups can test new AI systems in a controlled, government-approved environment without full compliance, as long as risks are contained)

  51. Government funding for compliance assistance for SMEs (Government helps small businesses pay for compliance (e.g., audits, safety testing)).

  52. 18-month transition for existing systems; 6-month for new entrants

  53. Recognition of EU/UK/APEC compliance certifications (If a company already meets EU, UK, or APEC AI standards, Australia will recognise those certifications to avoid duplication of costs, effort and makes Australia “easy to do business with.”)

  54. The Joint Selection Committee should recommend these measures be enacted where possible before the proposed early 2027 commencement of Australian Standards for AI legislation

  55. Rather than "last resort" AI-specific regulation, Australia should adopt at a minimum targeted mandatory guardrails for high-risk settings while preserving flexibility elsewhere: Tier 1 (High Risk): Mandatory  Healthcare diagnostic AI      Employment screening/termination Credit/lending decisions Criminal justice risk assessment Content generation with labelling Critical infrastructure control systems

  56. Tier 2 (Medium Risk): Conditional Mandatory   Automated customer service with human override Marketing personalisation with disclosure Tier 3 (Low Risk): Voluntary    Internal productivity tools Research/test environments Non-decision-support analytics

  57. Regular review cycles (2-year sunset provisions) to prevent regulatory ossification

Read the original

Open in the interactive guide