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 98

Centre for Law as Protection, Deakin University

Submitter group: Universities.

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.

Recommends changes to the Policy for the Responsible Use of AI in Government and the AI Impact Assessment Tool (defining dignitary harm, default privacy and security assessments), and reforms on privately owned AI surveillance, data aggregation, AI-generated image-based sexual abuse, human control in security decisions and agentic AI. Also asks for research funding and mandated safety by design.

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 defining the government AI Policy's harm threshold to include dignitary harms.

    Recommendation H.1: Update Appendix B of the Policy for the Responsible Use of AI in Government to include a definition of ‘harm’ to capture dignitary harms (see drafting above).

  2. Argues that automation bias may limit human review's ability to manage risk, and recommends removing 'without human review' from the threshold criteria.

    There is considerable research around automation bias indicating that humans are influenced by or deferential to technical processes. 4 This criterion should be amended to remove the words ‘without human review’, because such review may do little to manage risk.

  3. Recommends a binding 'meaningful human control' standard specifying minimum requirements, such as reviewer access to underlying evidence, adequate review time and genuine authority to override.

    Legislate a binding ‘meaningful human control’ standard. The standard should specify minimum requirements for what counts as ‘meaningful’ — e.g. the human reviewer must have access to underlying evidence (not just a confidence score), adequate time to review, and genuine authority to override.

  4. Argues that Australia's legal framework offers no clear rules governing privately owned AI-powered smart technologies, and that this gap must be addressed with a meaningful regulatory regime.

    Despite the stakes for privacy, dignity and identity, Australia’s legal framework offers no clear rules governing these privately owned AI-powered smart technologies. This legal gap must be addressed with a meaningful regulatory regime.

  5. Argues that aggregation of low-sensitivity data can build a detailed picture of a person while the law regulates each data point in isolation.

    individually low-sensitivity data points can be aggregated into a detailed, continuous picture of a person’s movements, transactions, health, relationships and communications. The law regulates each data point in isolation; it does not yet grapple with what the aggregate reveals (or has potential to reveal).

  6. Recommends mandating safety by design for AI tools on the Australian market capable of generating non-consensual intimate imagery.

    Mandate ‘safety by design’ approaches to the design, development and deployment of AI tools on the Australian market capable of generating non-consensual intimate imagery

Recommendations

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

  1. Define harm to include AI-incident consequences or any other non-trivial adverse effect on an individual or group arising from AI analysis or use of their data.

    Harm includes each of the consequences in the definition of ‘AI incident’ or any other non-trivial adverse effect on an individual or group arising from AI’s analysis or use of data relating to them.

  2. Remove 'without human review' and 'more than insignificant' from the Appendix C criteria of the Policy; remove 'significantly' or replace it with 'non-trivially'.

    Remove the words ‘without human review’ and ‘more than insignificant’ from the criteria in Appendix C of the Policy. The word ‘significantly’ should be removed or replaced with ‘non-trivially’.

  3. Require default Privacy Impact Assessments and Security Risk Assessments under Part 7 of the AI Impact Assessment Tool, with commitments to comply with all recommendations, mitigations and findings.

    Amend Part 7 of the AI Impact Assessment Tool to require that both a Privacy Impact Assessment and a Security Risk Assessment be conducted as a default practice, with the questions re-designed to require a commitment to comply with any recommendations, mitigations or findings of each assessment.

  4. Review and update the PSPF Guidelines so that all 'personal information' has a default rating of OFFICIAL: Sensitive and is handled accordingly.

    Review and update the PSPF Guidelines to indicate that all ‘personal information’ should have a default rating of OFFICIAL: Sensitive and be handled accordingly.

  5. Amend the Privacy Act and/or surveillance legislation to impose obligations on private actors using camera, audio or sensor data collection in public or semi-public spaces, including express (not implied) consent before capture, retention or sharing.

    Amend the Privacy Act, surveillance legislation or both to impose specific obligations on private actors deploying camera-, audio- or sensor-based data collection in public/semi-public spaces, including a requirement for express (not implied) consent before an individual’s data is captured, retained or shared.

  6. Provide clear and effective redress for breaches involving privately owned surveillance devices.

    Effective governance should include limits on privacy invasions by privately-owned surveillance devices, including limiting the tracking or capturing of individuals without express consent, and providing clear and effective redress mechanisms for breaches.

  7. Commission a review of the statutory privacy tort's operation within 2 years of commencement, testing whether its consent defence, 'misuse' standard and fault threshold capture data-aggregation harms.

    Commission a review of the statutory tort’s operation within 2 years of commencement, specifically testing whether its consent defence, undefined ‘misuse’ standard, and fault threshold adequately capture data-aggregation harms.

  8. Fund empirical, qualitative and quantitative research on diverse lived experiences of AI-generated image-based sexual abuse, including nationally representative victimisation and perpetration prevalence studies.

    Fund empirical, qualitative and quantitative research on diverse lived experiences of AI- IBSA, including nationally representative prevalence (victimisation and perpetration) studies.

  9. Fund expert-led, evidence-based, age-tailored AI-IBSA education and prevention grounded in consent and respect, regularly evaluated and refined, within holistic gender-based-violence responses including Australia’s National Plan to End Violence against Women and Children 2022–2032.

    Fund AI-IBSA-specific, expert-led and evidenced-based education and prevention initiatives, which are tailored to different age groups, underpinned by principles including consent and respect, and regularly evaluated and refined; and (ii) situate and implement these initiatives as part of broader holistic responses to gender-based violence, including Australia’s National Plan to End Violence against Women and Children 2022–2032.

  10. Do more to regulate access to or availability of AI-IBSA tools in Australia.

    more could be done in Australia to regulate access to or the availability of AI-IBSA tools (eg, akin to regulatory approach adopted by the City Attorney of San Francisco), 42 as well as to mandate safety by design approaches to designing, developing and/or deploying such tools.

  11. Mandate safety by design in the design, development and deployment of AI tools on the Australian market capable of generating non-consensual intimate imagery.

    Mandate ‘safety by design’ approaches to the design, development and deployment of AI tools on the Australian market capable of generating non-consensual intimate imagery

  12. Include general-purpose tools that foreseeably enable such use, building on the eSafety Commissioner’s existing safety by design initiatives.

    purpose tools that foreseeably enable such use, building on the eSafety Commissioner’s existing safety by design initiatives.

  13. Audit human-cognition standards in Commonwealth and military law for AI-assisted defence, national-security and law-enforcement decisions; report to Parliament which are unfit for machine application and which technical concepts replace them in practice.

    Audit Commonwealth and military law for standards premised on human cognition (good faith, reasonable belief, reasonable suspicion, acceptable doubt) as applied to AI-assisted decisions in defence, national security and law enforcement, and report to Parliament on which standards are unfit for machine application and which technical concepts replace them in practice.

    A program corrected obvious text-recognition errors in this quote for display.

  14. Legislate a binding 'meaningful human control' standard specifying minimum requirements, such as reviewer access to underlying evidence, adequate time and genuine authority to override.

    Legislate a binding ‘meaningful human control’ standard. The standard should specify minimum requirements for what counts as ‘meaningful’ — e.g. the human reviewer must have access to underlying evidence (not just a confidence score), adequate time to review, and genuine authority to override.

  15. Create an enforceable right for those wrongly flagged by a national-security or critical-infrastructure AI system to be notified where operationally feasible and seek independent review and correction, with a statutory response timeframe.

    Create an enforceable right for individuals and entities wrongly flagged by a national- security or critical-infrastructure AI system to be notified where operationally feasible and to seek independent review and correction, with a statutory response timeframe.

    A program corrected obvious text-recognition errors in this quote for display.

  16. Strengthen accountability through individual responsibility structures for decisions to deploy AI-powered decision systems and organisational responsibility frameworks for AI harms, including redress for victims and internal harm mitigation.

    Strengthen accountability through both individual responsibility structures (concerning decisions to deploy AI-powered decision systems) and through organisational responsibility frameworks (for AI harms). Acknowledging that individual responsibility is not always meaningful in complex and highly technical systems, there is an acute need to develop stronger accountability mechanisms focused on organisational responsibility and effective redress. We recommend developing a framework for organisational responsibility that provides mechanisms for redress (to victims) and for future improvement and harm mitigation (internally).

    A program corrected obvious text-recognition errors in this quote for display.

  17. Commission a beyond-Richardson surveillance review covering exemptions, immunities and ancillary-law data-sharing provisions, not just explicit surveillance powers; design safeguards and oversight for AI-enabled data collection and aggregation tools.

    Commission a ‘beyond-Richardson’ review of Australia’s surveillance architecture, to document and examine gaps created by exemptions, immunities and ancillary-law data-sharing provisions (not just explicit surveillance powers), and design adequate safeguards and oversight mechanisms for AI-enabled data collection and aggregation tools.

    A program corrected obvious text-recognition errors in this quote for display.

  18. Harmonise national-security/law-enforcement authorisation thresholds for AI-enabled data access or repurposing to resulting inference’s intrusiveness, rather than the source law’s label or acquisition method, closing the threshold-shopping gap.

    Harmonise authorisation thresholds for AI-enabled access to, or repurposing of, data for national-security or law-enforcement purposes to the intrusiveness of the resulting inference — not the statutory label of the source law or the technical method used to obtain it — closing the threshold-shopping gap Krebs and Mann document in existing law.

    A program corrected obvious text-recognition errors in this quote for display.

  19. Consider direct legal duties on AI agents, whether government agencies wielding state power should use them, and opt-in rather than opt-out rules for personalisation.

    Consider whether this necessitates directly imposing legal duties on AI agents;>! whether AI agents can and should be used by government agencies wielding state power and whether rules should be in place about opting in rather than out of personalisation.*?

    A program corrected obvious text-recognition errors in this quote for display.

  20. Examine major technology companies’ AI personalisation proposals and consider how laws can ensure guardrails.

    Examine recent proposals by the major technology companies in relation to AI personalisation and consider the ways in which laws can be crafted to ensure guardrails are in place

    A program corrected obvious text-recognition errors in this quote for display.

  21. Consider restricting or prohibiting opt-out arrangements for AI technologies or tools embedded within or offered by digital platforms, addressing present and likely future developments.

    present and likely future developments of the technologies (eg, restricting or prohibiting ‘opt out’ arrangements for AI technologies/tools embedded within or offered by digital platforms).

    A program corrected obvious text-recognition errors in this quote for display.

  22. Consider amendments reflecting Australia's human rights obligations in AI regulation and expert recommendations such as the Australian Human Rights Commission's Human Rights and Technology report.

    Consider amendments to reflect Australia’s human rights obligations in relation to the regulation of AI and of expert recommendations, such as the recommendations of the Australian Human Rights Commission’s report on Human Rights and Technology.

    A program corrected obvious text-recognition errors in this quote for display.

Read the original

In some quotes a program corrected obvious text-recognition errors for display.

Open in the interactive guide