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 502

Mr Craig Weniton

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.

Argues that AI makes existing harms cheaper and that targeted reforms should close coverage, enforcement and measurement gaps. Proposes safeguards for scams, deepfakes, children, automated decisions, critical infrastructure and elections, alongside stronger AI Safety Institute powers, lobbying transparency and parliamentary oversight. Distinguishes documented harms from emerging scenarios requiring investigation; recommends adaptable, scalable measures and prioritises nine numbered asks.

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 criminal AI tools lower the skill, time and cost needed for scams, intrusions and abuse rather than raising the capability ceiling.

    The tools now in criminal hands do not raise the ceiling of what AI can do; they lower the floor of who can do it. A scam, an intrusion or a piece of abuse material that once required skill, time and money now requires a subscription and an afternoon.

  2. Argues that Australia's main weaknesses are gaps in coverage, sequencing, enforcement capacity and measurement rather than an absence of regulatory architecture.

    Australia has built substantial AI-relevant law and institutions. The most immediate weaknesses identified in this submission are gaps in coverage, sequencing, enforcement capacity and measurement rather than a complete absence of regulatory architecture.

  3. Says the Scams Prevention Framework excludes non-bank payment providers, crypto exchanges, email and VoIP.

    Gap K1: The SPF does not cover the channels AI scams run through. Treasury's own consultation drew submissions arguing that non-bank payment providers, cryptocurrency exchanges, email services and VoIP providers form the backbone of how scams are initiated and monetised, yet none are designated.

  4. States that the 2024 Commonwealth deepfake offences are confined to sexual material, and that creating a voice clone without consent is not in itself unlawful.

    Creating a voice clone of a person without consent is not, in itself, unlawful. The 2024 Commonwealth deepfake offences are confined to sexual material.

  5. Argues that child-safety codes address what a chatbot says but not how it is designed.

    The Age-Restricted Material Codes are about what a chatbot will say. The harms documented by eSafety and by the International AI Safety Report 2026 also arise from how a chatbot is designed: sycophancy (agreeing with the user), engagement-maximising persistence, simulated intimacy, memory of personal disclosures, and 'streak' or notification mechanics that make a companion feel like a relationship.

  6. Argues strategic resilience means managed dependence, requiring visibility, domestic evaluation and defence, continuity plans and legal mechanisms to act.

    Strategic resilience should be treated as managed dependence: Australia needs visibility into critical AI dependencies, domestic evaluation and defensive capability, continuity planning, and legal mechanisms that allow government to act when a dependency becomes unsafe or unavailable.

Recommendations

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

  1. Designate non-bank payments, crypto exchanges, email/VoIP and consumer synthetic-media providers under the SPF; require verified consent for real-person cloning, output watermarking and fixed-timeframe takedown responses.

    Designate non-bank payment providers, cryptocurrency exchanges, email and VoIP providers under the SPF, and create a new designated category for providers of consumer synthetic-media generation services (voice cloning, face- swap, AI advertising generation), with obligations to verify consent for cloning real people, watermark outputs, and respond to takedown requests within fixed timeframes.

  2. Create a technology-neutral Commonwealth offence and a civil right of action with a fast-track eSafety takedown pathway for synthetic likenesses made without consent with intent to deceive, defraud, harass or cause serious harm, and for reckless commercial provision of tools without reasonable safeguards.

    Create a technology- neutral Commonwealth offence, and a civil right of action with a fast-track takedown pathway through eSafety, for creating or distributing a synthetic likeness (voice, face or body) of a real person without consent with intent to deceive, defraud, harass or cause serious harm, and for reckless commercial provision of tools without reasonable safeguards.

  3. Restrict nudify/undressing services, with app-store and search-engine delisting, obligations on payment processors and hosts, and rapid re-designation after name/ownership changes.

    Enact the power to restrict the availability of nudify and undressing services, with obligations on app stores, search engines, payment processors and hosting providers, and a rapid re-designation mechanism for services that relaunch under new names or ownership.

  4. Require app-store and search-engine delisting when restricting nudify and undressing services.

    Legislate the promised power to restrict 'nudify' and undressing services, with app-store and search-engine delisting obligations, so that enforcement stops being a whack-a-mole exercise against relaunched services.

  5. Direct agencies to retire voice-only biometric authentication or pair it with a second factor by a fixed date; ask APRA and ASIC to require the same of regulated financial institutions; and publish a national standard, maintained by the AI Safety Institute and ASD, for biometric authentication resilience to synthetic media.

    Direct Commonwealth agencies to retire voice-only biometric authentication or pair it with a second factor by a fixed date; ask APRA and ASIC to require the same of regulated financial institutions; and publish a national standard, maintained by the AI Safety Institute and ASD, for the resilience of biometric authentication to synthetic media.

  6. Ask ACMA and the telecommunications industry, under the SPF telecommunications code, to make carrier-generated or synthetic voicemail greetings the default, offer a text-only or generic greeting option prominently, and prompt existing customers with personalised greetings about cloning risk.

    Ask ACMA and the telecommunications industry, under the SPF telecommunications code, to make carrier-generated or synthetic greetings the default for voicemail, to offer customers a text-only or generic greeting option prominently, to prompt existing customers with personalised greetings about the cloning risk,

  7. Include voice-sample hygiene (voicemail, public video, 'safe words') in the National Anti-Scam Centre's public education.

    to include voice-sample hygiene (voicemail, public video, 'safe words') in the National Anti-Scam Centre's public education.

  8. Legislate the digital duty of care so it expressly applies to general-purpose chatbots, AI companions and design features, with a specific obligation to design age-appropriate experiences for minors.

    Legislate the duty of care accepted in the Government's April 2026 response, drafted so that it expressly applies to general-purpose chatbots and AI companions and to design features (sycophancy, engagement- maximisation, simulated intimacy, memory and persistence), with a specific obligation to design age-appropriate experiences for minors.

  9. Require generative-AI services accessible to minors to implement crisis detection, referral to Australian support services and human escalation; have the AI Safety Institute red-team these protocols annually and publish results; and mandate age assurance at app-store and operating-system level.

    Require any generative-AI service accessible to Australian minors to implement crisis detection, referral to Australian support services and human escalation; task the AI Safety Institute with red-teaming these protocols annually and publishing results; and mandate age assurance at the app-store and operating-system layer.

  10. Legislate the Robodebt Royal Commission framework: a register of automated and AI-assisted decision systems in government, a statutory right to human review of significant decisions, assessor and reviewer power to override automated outcomes, independent algorithm audit and plain-language explanation, applied to aged care and the NDIS first.

    Legislate the framework the Robodebt Royal Commission recommended: a register of automated and AI- assisted decision systems in government; a statutory right to human review of any decision with significant effect; a requirement that assessors and reviewers can override automated outcomes; independent algorithm audit; and plain-language explanation. Apply it to aged care and the NDIS first.

  11. Establish a national AI-harm reporting scheme covering serious financial, psychological and safety harms to individuals, with a public register.

    Establish a national AI-harm incident reporting scheme covering serious harms to individuals (financial, psychological, safety), with a public register, so that the AI-enabled share of scam, abuse and welfare harm can be measured rather than guessed.

  12. Introduce a consistent 'AI-enabled' flag across Scamwatch, ReportCyber, IDCARE, AFCA and eSafety reporting, publish the data annually, and establish a single national reporting front door for AI-enabled fraud and impersonation.

    Introduce a consistent 'AI-enabled' flag across Scamwatch, ReportCyber, IDCARE, AFCA and eSafety reporting, publish the resulting data annually, and establish a single national reporting front door for AI-enabled fraud and impersonation.

  13. Fund in-language, culturally appropriate AI-scam prevention for CALD and First Nations communities, aged-care and seniors' organisations and small business, and require designated SPF entities to report on outcomes for these groups.

    Fund in-language, culturally appropriate AI-scam prevention for CALD and First Nations communities, aged-care and seniors' organisations, and small business, and require designated SPF entities to report on outcomes for these groups.

  14. Include people with disability in funded AI-specific scam and abuse prevention.

    Fund AI-specific scam and abuse prevention for the groups currently missing from the response: older Australians, culturally and linguistically diverse communities, First Nations communities, people with disability, and boys and young men targeted by sextortion

  15. Include boys and young men targeted by AI-enabled sextortion as a named priority group in prevention funding, with de-stigmatised reporting pathways and school-based education.

    Include boys and young men targeted by AI- enabled sextortion as a named priority group in prevention funding, with de-stigmatised reporting pathways and school-based education.

  16. Develop a national protocol for school response to AI-enabled abuse and AI incidents.

    Develop a national protocol for school response to AI-enabled abuse and AI incidents;

  17. Fund an Australian longitudinal study of children's conversational-AI use, with lawful researcher access to platform data.

    fund an Australian longitudinal study of children's conversational-AI use with lawful researcher access to platform data;

  18. Prohibit training on, or retention of, children's conversational data without a clear lawful basis.

    prohibit training on, or retention of, children's conversational data without a clear lawful basis.

  19. Task the Aged Care Quality and Safety Commission, the NDIS Commission and the TGA with binding standards for AI in care, including consent, human oversight and an adverse-event reporting scheme feeding the national AI-harm register.

    Task the Aged Care Quality and Safety Commission, the NDIS Quality and Safeguards Commission and the TGA with binding standards for AI in care settings, including consent, human oversight and an adverse-event reporting scheme feeding the national AI-harm register at Recommendation 8.

  20. Fund AI-harm legal assistance within existing community legal centres and IDCARE.

    Fund AI-harm legal assistance within existing community legal centres and IDCARE;

  21. Extend the Privacy Act to small businesses deploying AI on individuals.

    extend the Privacy Act to small businesses deploying AI on individuals;

  22. Give family-violence services and courts guidance and funding to recognise AI-enabled coercive control and synthetic evidence.

    and give family-violence services and courts guidance and funding to recognise AI- enabled coercive control and synthetic evidence.

  23. Have NASC, eSafety and IDCARE co-design voluntary Mindframe-style guidelines with the Press Council, MEAA and victim-support bodies: victim consent, default de-identification, omission of bank/timeline/contact/support details, and recovery-scam warnings.

    Task the National Anti-Scam Centre, eSafety and IDCARE to develop, with the Australian Press Council, the Media, Entertainment and Arts Alliance and victim-support organisations, voluntary 'safe scam reporting' guidelines on the Mindframe model — covering victim consent, de-identification by default, omission of operational specifics (bank, timeline, contact channel, support pathway) and a standard 'what happens next' warning about recovery scams;

  24. Apply the safe-reporting standard to Commonwealth agency/regulator releases and case studies and include recovery-scam warnings in every Scamwatch, bank and IDCARE victim interaction.

    require Commonwealth agencies and regulators to apply the same standard in their own media releases and case studies; build a recovery-scam warning into every Scamwatch, bank and IDCARE victim interaction;

  25. Fund research on repeat victimisation, including whether publicly identified victims are re-targeted at higher rates.

    fund research on repeat victimisation, including whether publicly identified victims are re-targeted at higher rates.

  26. Through the Children's Online Privacy Code, Privacy Act reforms and OAIC and eSafety guidance, establish reasonable-disclosure defaults for institutions publishing identifiable material, including opt-in, per-publication consent for children's and staff voices and images.

    Through the Children's Online Privacy Code, the Privacy Act tranche-two reforms and guidance from the OAIC and eSafety, establish reasonable-disclosure defaults for institutions that publish identifiable material about people who did not choose the publication: voice recordings and facial imagery of children and staff published on an opt-in, per-publication basis rather than under blanket consent;

  27. Require removal of published material when a student, resident or employee leaves, and an obligation to maintain an inventory of published identifiable material so it can be minimised and removed.

    removal of published material when a student, resident or employee leaves; an obligation to maintain an inventory of published identifiable material so that it can be minimised and removed;

  28. Clarify that the employee records exemption does not extend to material an employer publishes to the world.

    clarification that the employee records exemption does not extend to material an employer publishes to the world.

  29. Frame disclosure defaults as institutional duties on schools, employers, clubs and care providers, rather than families, with guidance against blanket enrolment-consent clauses that transfer liability without reducing exposure.

    These should be framed as defaults and duties on institutions — schools, employers, clubs and care providers — rather than on families, and the guidance should warn expressly against consent theatre: a further clause in an enrolment form transfers liability to the parent without reducing exposure.

  30. Require large platforms/generative-AI services to preserve content credentials and label generated media; have the Attorney-General’s Department and Council of Attorneys-General set an evidence-authentication protocol; fund courts/tribunals/police forensics.

    Require large platforms and generative-AI services to attach and preserve content credentials and to label AI-generated media; task the Attorney-General's Department and the Council of Attorneys-General with a national protocol for authenticating digital evidence; and fund forensic capacity for courts, tribunals and police.

  31. Align AI-media provenance and labelling requirements with international standards.

    10. Introduce content-provenance and labelling requirements for AI-generated media on large platforms, aligned with international standards, together with rules for the authentication of digital evidence in Australian courts.

  32. Define AI systems under SOCI; require critical-asset AI inventories, weight integrity/provenance, injection/exfiltration controls, behavioural-drift monitoring, human override and AI-failure incident reporting in Risk Management Programs.

    Amend the Security of Critical Infrastructure Act and rules to define 'AI system' and to require, within the Risk Management Program, an inventory of AI systems in critical assets, AI-specific risk controls (integrity of weights, provenance, prompt-injection and data-exfiltration controls, monitoring for behavioural drift, human-override), and incident reporting for AI-specific failures.

  33. Explicitly include model-hosting infrastructure operating critical assets in SOCI’s scope.

    Amend the Security of Critical Infrastructure Act 2018 to make clear that AI systems and model- hosting infrastructure used to operate critical infrastructure assets are within scope, and require AI- specific risk controls in the Risk Management Program.

  34. Amend SOCI rules to require systems of national significance and other high-criticality assets to conduct periodic intelligence-informed threat hunting for living-off-the-land persistence, with ASD support and AI-enabled behavioural detection.

    Amend SOCI rules to require systems of national significance and other high-criticality assets to conduct periodic intelligence-informed threat hunting for living-off-the-land persistence (with ASD support and AI-enabled behavioural detection under Recommendation 15),

  35. Amend SOCI rules to require systems of national significance and other high-criticality assets to report suspected pre-positioning to ASD even where no disruption has occurred.

    Amend SOCI rules to require systems of national significance and other high-criticality assets to conduct periodic intelligence-informed threat hunting for living-off-the-land persistence (with ASD support and AI-enabled behavioural detection under Recommendation 15), to report suspected pre-positioning to ASD even where no disruption has occurred,

  36. Amend SOCI rules to require systems of national significance and other high-criticality assets to demonstrate a tested capability to isolate operational technology for 90 days.

    Amend SOCI rules to require systems of national significance and other high-criticality assets to conduct periodic intelligence-informed threat hunting for living-off-the-land persistence (with ASD support and AI-enabled behavioural detection under Recommendation 15), to report suspected pre-positioning to ASD even where no disruption has occurred, and to demonstrate a tested capability to isolate operational technology for 90 days;

  37. Fund a shared detection service for regional and smaller operators who cannot build their own.

    and fund a shared detection service for regional and smaller operators who cannot build their own.

  38. Create a standing regime in which Home Affairs, advised by ASD and the AI Safety Institute, assesses AI models and services for foreign ownership, control and influence risk and can direct critical-infrastructure entities not to use, or to apply conditions to, specified models, including open-weight models run locally.

    Convert the current ad hoc approach (PSPF Direction 001-2025) into a standing regime under which the Department of Home Affairs, advised by ASD and the AI Safety Institute, assesses AI models and services for foreign ownership, control and influence risk and can direct critical-infrastructure entities not to use, or to apply conditions to, specified models — with attention to open-weight models that can be run locally and therefore evade website- and app-level bans.

  39. Through the Australian Standards for AI or SPF designation, require AI providers serving Australian users to maintain abuse detection, revoke access on detection, report serious misuse to ASD within fixed timeframes and share abuse indicators.

    Through the Australian Standards for AI, or by designating AI model and API providers under the SPF, require providers serving Australian users to maintain abuse detection for cybercrime and fraud, to revoke access on detection, to report serious misuse to ASD within fixed timeframes, and to share jailbreak and abuse indicators with regulators and designated entities;

  40. Review Part 10.7 of the Criminal Code so operating, marketing or reselling an AI service designed or promoted for fraud, social engineering or cybercrime is clearly an offence, with sales channels brought within the SPF's reach.

    review Part 10.7 of the Criminal Code so that operating, marketing or reselling an AI service designed or promoted for fraud, social engineering or cybercrime is clearly an offence, with the sales channels used for such services brought within the SPF's reach.

  41. Establish a confidential register, maintained by the Cyber and Infrastructure Security Centre, of AI systems and large-scale compute in critical infrastructure and government, and a mandatory pre-deployment notification regime for agentic AI in operational-technology environments.

    Establish a confidential register, maintained by the Cyber and Infrastructure Security Centre, of AI systems and large-scale compute deployed in critical infrastructure and government, and a mandatory pre-deployment notification regime for agentic AI in operational-technology environments.

  42. Give the AEC the legislative/technical tools it lacks; enact a federal electoral-deepfake offence modelled on SA/NSW and progress truth-in-political-advertising law before the next federal election.

    Give the AEC the legislative tools and technical capability it has identified it lacks; enact a federal electoral deepfake offence modelled on South Australia and New South Wales; and progress federal truth-in-political-advertising law ahead of the next federal election.

  43. Fund a national defensive AI program led by ASD's ACSC with the AI Safety Institute, applying AI-assisted vulnerability discovery, patch prioritisation and threat-hunting across Commonwealth systems and critical-infrastructure operators, with a subsidised tier for small and medium entities in critical supply chains and an annual public account of vulnerabilities found and fixed.

    Fund a national program — led by ASD's ACSC with the AI Safety Institute — to apply AI-assisted vulnerability discovery, patch prioritisation and threat-hunting across Commonwealth systems and critical-infrastructure operators, with a subsidised service tier for small and medium entities in critical supply chains, and publish an annual account of vulnerabilities found and fixed.

  44. Establish ASD ACSC/AFP-led JPC3/AISI criminal-AI monitoring, assess capability against Australian defences, share indicators with SPF entities/international partners, and report annually to Parliament in unclassified form.

    Establish a standing function, jointly staffed by ASD's ACSC, the AFP-led Joint Policing Cybercrime Coordination Centre and the AI Safety Institute, to monitor malicious AI services, jailbreak-as-a-service operators, poisoned-model distribution and AI-embedded malware; to assess their capability against Australian defences; to share indicators with SPF-designated entities and international partners; and to report annually to Parliament in unclassified form.

  45. Apply foreign-investment review, physical and cyber security standards, workforce vetting and grid-resilience obligations to large AI data centres as a condition of approval, and designate sovereign compute capacity in Australian-controlled facilities reserved for government and critical-infrastructure workloads in a crisis.

    Apply foreign- investment review, physical and cyber security standards, workforce vetting and grid-resilience obligations to large AI data centres as a condition of approval; and designate sovereign compute capacity, in Australian-controlled facilities, reserved for government and critical-infrastructure workloads in a crisis.

  46. Create an Australian assurance scheme for AI components used in critical infrastructure, building on ASD's existing evaluation programs.

    Create an Australian assurance scheme for AI components used in critical infrastructure (building on ASD's existing evaluation programs),

  47. Require critical operators to plan for provider withdrawal, degradation or compromise, and extend a subsidised AI-assisted security service to small suppliers in critical supply chains.

    require critical operators to plan for provider withdrawal, degradation or compromise, and extend a subsidised AI-assisted security service to small suppliers in critical supply chains (see Recommendation 15).

  48. Establish the AI Safety Institute in legislation with a defined mandate, information-gathering and model-access powers, a guaranteed multi-year appropriation well above the current $29.9 million, independence in its published findings and an annual report to Parliament.

    Establish the Australian AI Safety Institute in legislation with a defined mandate, information-gathering and model-access powers, a guaranteed multi-year appropriation substantially above the current $29.9 million, independence in its published findings, and an annual report to Parliament.

  49. Require AISI notification before frontier models become materially available to Australians through local or overseas services; define updateable capability/compute thresholds through transparent technical instruments, capturing material post-deployment capability gains.

    Require developers to notify AISI before making a frontier model materially available to Australian users, whether through a locally operated service or an overseas-hosted API or consumer service. Define 'frontier model' by capability and/or compute thresholds set in legislation but updateable by AISI through a transparent technical instrument, with a mechanism to capture material capability increases after deployment.

  50. Require secure pre-deployment evaluation access under statutory confidentiality protections for model weights, trade secrets and security-sensitive findings.

    Require secure pre-deployment evaluation access under statutory confidentiality protections for model weights, trade secrets and security-sensitive findings.

  51. Specify proportionate consequences for non-compliance with pre-deployment notification, and a pathway by which AI Safety Institute findings can inform deployment conditions, public advisories or regulator referrals.

    The regime should specify proportionate consequences for non-compliance and a clear pathway by which AISI findings can inform deployment conditions, public advisories or referrals to the regulator with relevant statutory powers.

  52. For open-weight releases where pre-deployment access cannot control downstream copies, require advance notification, pre-public-release evaluation for developers within Australian jurisdiction, risk-information publication and post-release monitoring of material misuse and capability changes.

    Open-weight releases require a separate treatment: where pre-deployment access cannot practically control downstream copies, the obligation should focus on advance notification, evaluation before public release where the developer is within Australian jurisdiction, publication of risk information, and post- release monitoring of material misuse and capability changes.

  53. Establish a whole-of-government AI capability-monitoring and horizon-scanning function led by ONI with ASD, ASIO and AISI, producing a classified annual capability assessment and unclassified public summary, and a standing joint parliamentary committee on AI with powers analogous to the PJCIS to receive it.

    Establish a whole-of-government AI capability-monitoring and horizon-scanning function led by ONI with ASD, ASIO and AISI, producing a classified annual capability assessment and an unclassified public summary; and create a standing joint parliamentary committee on artificial intelligence with powers analogous to the Parliamentary Joint Committee on Intelligence and Security to receive it.

  54. Extend the Lobbying Code of Conduct and Register of Lobbyists to in-house lobbyists and industry associations, and place the Code on a legislative footing with penalties.

    Extend the Lobbying Code of Conduct and the Register of Lobbyists to in-house lobbyists and industry associations, and place the Code on a legislative footing with penalties,

  55. Require publication of ministerial and senior-official diaries for meetings on AI policy, beginning with the Office of AI, DISR and the Attorney-General's Department.

    require publication of ministerial and senior-official diaries for meetings concerning AI policy, beginning with the Office of AI, the Department of Industry, Science and Resources and the Attorney-General's Department;

  56. Require memoranda of understanding and investment commitments negotiated with AI developers to be tabled, with redactions confined to genuinely commercial terms.

    require that memoranda of understanding and investment commitments negotiated with AI developers be tabled, with commercial-in-confidence redactions confined to genuinely commercial terms;

  57. Ask the Attorney-General's Department to report to Parliament on whether the Foreign Influence Transparency Scheme reaches policy advocacy by Australian subsidiaries of foreign AI developers.

    ask the Attorney-General's Department to report to Parliament on whether the Foreign Influence Transparency Scheme, as applied, reaches policy advocacy by the Australian subsidiaries of foreign AI developers.

  58. Fund a five-year program placing technical AI capability in each AI-relevant regulator, with AISI as shared centre of expertise, ASD secondments and a common evidence base, and require each regulator to report annually on AI-related enforcement and capability.

    Fund a five-year program to place technical AI capability (evaluators, data scientists, forensic specialists) in each AI-relevant regulator, with AISI as the shared centre of expertise, secondments from ASD, and a common evidence base; and require each regulator to report annually on AI-related enforcement and capability.

  59. Reinstate mandatory high-risk guardrails through the 2027 Australian Standards for AI covering eligibility, credit, insurance, employment, health and care.

    Revisit the shelved mandatory guardrails for high-risk AI so that the Australian Standards for AI planned for 2027 govern AI used on Page 18 of 40 Submission — Joint Select Committee on Artificial Intelligence — Terms of Reference (k), (l), (m) people in eligibility, credit, insurance, employment, health and care — not only the environmental footprint of infrastructure.

  60. Reconsider whether scam victims should bear losses from technically flawless impersonations without mandatory reimbursement.

    Gap K4: No mandatory reimbursement. The SPF creates dispute-resolution pathways but does not mandate reimbursement of victims in the way the United Kingdom's authorised push payment reimbursement regime does. As AI makes scams harder to detect, the fairness of leaving losses with the individual who was deceived by a technically flawless impersonation deserves reconsideration.

  61. Independently test Digital ID/myGov security assumptions against current generative AI and report to Parliament.

    As Australia's Digital ID system and myGov become the front door to government services, the incentive to defeat their onboarding and recovery flows with AI-generated documents, faces and voices rises accordingly. The security assumptions of these systems should be tested by an independent body against current generative capability, and the results reported to Parliament.

  62. Introduce statutory takedown timeframes, a hash-matching scheme for known deepfake abuse material analogous to that used for child abuse material, and funded victim-support services.

    Statutory takedown timeframes, a hash-matching scheme for known deepfake abuse material analogous to that used for child abuse material, and funded victim-support services would materially reduce harm.

  63. Prohibit retention of children’s conversational data beyond a session, alongside unlawful training use.

    The Children's Online Privacy Code will help; a clear prohibition on training on children's data without a lawful basis, and on retaining children's conversational data beyond the session, would help more.

  64. Prioritise Recommendations 1, 2, 3, 5, 7, 11, 12, 17 and 18 if only a few can proceed this Parliament.

    Sequencing. If only a few of these can be progressed in this Parliament, I would urge the Committee to prioritise items 1, 2, 3, 5, 7, 11, 12, 17 and 18. They are the measures that close the widest gaps at the lowest cost, and most of them build on frameworks that already exist.

  65. Check existing protections against near-zero-cost AI-enabled harm and adjust those that no longer hold.

    The protections do not need to be reinvented, but each of them needs to be checked against the question — does this still hold when the cost of the harm falls to nearly zero? —and adjusted where the answer is no.

  66. Judge every recommendation against adaptability and scalability as AI advances.

    The second is that the speed of advance means every recommendation has to be judged against two tests: is it adaptable, and does it scale?

Read the original

Open in the interactive guide