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 257

Good Ancestors

Submitter group: Civil society.

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 Australia needs AI influence, frontier-risk expertise and targeted regulation. Recommends hosting training with national-interest conditions, copyright pathways, transparency, incident reporting, AISI funding, gene screening and agent liability reform. Its optional permit-and-creator-fund consultation model includes enforceable opt-outs; attached feedback says rights-holder groups preferred licensing and raised fund-certainty and treaty concerns, prompting proposed safeguards.

Key points

Quotes are the submitter’s own words. Text taken from documents, including scanned pages read by text-recognition software, can differ from the PDF. Check the original. Listed in the same order as on the submission’s page in the interactive guide.

  1. Argues that AI harms are already occurring and are warning shots of worse to come, and that Australia is exposed to these risks and unprepared.

    These harms are already occurring, and are warning shots of worse to come. Australia is exposed to these risks, and unprepared for them.

  2. Argues that hosting frontier AI training may be the most valuable position Australia could fill, strengthening its position to negotiate model access and oversight.

    Hosting frontier AI training may be the most valuable position Australia could fill. This would strengthen Australia's position to negotiate for continued access to frontier models and oversight of how they are developed.

  3. Argues that copyright is a barrier to hosting AI training and inference in Australia and that Government should develop an Australian approach that supports creators.

    Hosting AI training and inference in Australia might be the best way for Australia to secure a valuable position in the AI value chain and protect itself from the broader risks of AI. Copyright is a barrier to both.49

  4. Recommends that the AI Safety Institute be properly funded and build leading capability in an area such as compute verification.

    Australia should make its AI Safety Institute world-leading by ensuring it is properly funded and making meaningful contributions to one or more critical areas of technical AI safety, such as compute verification.

  5. Recommends mandating gene synthesis screening to reduce AI-enabled biosecurity risk, saying the Department of Agriculture already has the power to act.

    Government should mandate gene synthesis screening to reduce AI-enabled biosecurity risk. The Department of Agriculture already has the power to act under the Biosecurity Act 2015, at near-zero cost.

  6. Recommends that Government establish who is liable when AI agents cause harm.

    Government should establish who is liable when AI agents cause harm.

Recommendations

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

  1. Acknowledge in the Committee's report that AI poses catastrophic and potentially existential risks and that a proportionate government response is warranted.

    The Committee’s report should acknowledge that AI poses catastrophic and potentially existential risks, and that a proportionate government response is warranted.

  2. Have the Australian Government explicitly acknowledge catastrophic risks including cyberattacks on critical infrastructure, biological weapons development and loss of human control.

    AI capability is advancing rapidly, and greater capability brings greater risks—from cyberattacks on critical infrastructure to biological weapons development and the loss of human control. The Australian Government should explicitly acknowledge these risks as part of its response.

  3. Adopt the policy settings necessary to attract frontier AI training compute to Australia.

    Australia should adopt the policy settings necessary to attract frontier AI training compute to Australia.

  4. Seek model access at three levels: Australian public access equivalent to the US, restricted-model access for critical-infrastructure operators and other defenders, and government access equivalent to the US government for national security.

    Australia should seek access at three levels: 1) market access for the Australian public, equivalent to the US, 2) access to restricted or sensitive models for critical infrastructure operators and other defenders, and 3) Government access, equivalent to the US government for national security purposes.

  5. Secure oversight of AI development in Australia, including pre-deployment testing of Australian-trained models by the AI Safety Institute, disclosure of capabilities and risks, and reporting of critical safety incidents.

    Oversight over AI development in Australia, including pre-deployment testing of models trained in Australia by the Australian AI Safety Institute, disclosure of model capabilities and risks, and reporting of critical safety incidents to Government.

  6. Seek research and innovation compute on favourable terms, for example for Australian researchers, not-for-profits and start-ups.

    A portion of compute allocated for research and innovation on favourable terms (e.g., to Australian researchers, not-for-profits, and start-ups).

  7. Potentially require Australian-based AI engineers/researchers, high data-centre security standards, notification of large Australian training runs and fair contributions through taxation and other mechanisms.

    Other potential terms could include requiring some AI engineers and researchers to be based in Australia, ensuring AI data centres have high security standards, notifying Government of large training runs conducted in Australia, and ensuring fair contributions are made via taxation and other mechanisms.

  8. Measure Australian inference compute supply and demand on an ongoing basis.

    Measure Australian inference compute supply and demand on an ongoing basis.

  9. Set a target for Australia to become a net exporter of inference compute, and use it to guide energy supply, grid connection and planning decisions.

    Set a target for Australia to become a net exporter of inference compute, and use the target to guide decisions on energy supply, grid connection timelines, and planning approvals.

  10. Recommend an optional copyright pathway that makes AI training feasible without licensing every global work.

    An optional pathway that makes training feasible by not requiring licensing for every global work.

  11. Make opt-outs practical using the best available technology, such as robots.txt and content ID systems.

    Opt-out should be practical, including using the best available technology, such as robots.txt and content ID systems.

  12. Progress an Australian copyright approach, not a text and data mining exception, allowing training in Australia while supporting rights holders.

    Government should progress an Australian approach—not a text-and-data-mining exception—that supports rights holders while allowing training to occur in Australia.

  13. Proposes, as one potential model, an AI Training Permit for training in Australia, with a tiered fee, practical opt-out for rights holders, compliance with National Data Centre Expectations and other national-interest conditions.

    An AI Training Permit would be available for AI training in Australia. Obligations would include paying a tiered fee into consolidated revenue, granting practical consent and control to rights holders via an opt-out mechanism, complying with the National Data Centre Expectations, and meeting other national-interest conditions.

  14. Require developers of the most powerful models to provide technical descriptions and capabilities, dangerous-capability assessments and mitigations, and governance frameworks explaining compliance with safety standards.

    Transparency means requiring developers of the most powerful models to disclose what their models can do and the risks they carry. In practice, this means a technical description of the model and its capabilities, an assessment of its dangerous capabilities and what the company has done to address them, and a governance framework describing how the developer is meeting safety standards.

  15. Legislate critical-safety-incident definitions, notification periods and developer duties to report incidents to the Australian Government.

    Australia needs legislation to define what counts as a critical safety incident, set a notification period, and require companies to report this information to the Australian Government.

  16. Increase AISI funding to at least $50 million annually plus compute access.

    To satisfy its mandate, AISI funding should increase to at least $50 million per year plus compute access, roughly matching the UK's world-leading AISI when adjusting for population or GDP.58

  17. Direct the AI Safety Institute to build world-leading capability, such as compute verification, in areas that give allies and Australian regulators the information they need.

    Direct the Institute to build world-leading capability in areas that give allies and Australian regulators the information they need, such as compute verification.

  18. Require synthetic nucleic acids to come only from providers screening dangerous sequences and verifying customers, using the Director of Biosecurity's existing Biosecurity Act powers.

    Under the Biosecurity Act 2015, the Director of Biosecurity could require that synthetic nucleic acids come only from providers that screen orders for dangerous sequences and verify customers.

  19. Require imported synthetic nucleic acids to come from providers that screen orders, via import permit conditions.

    Add a condition to import permits in the Biosecurity Import Conditions system (BICON) requiring that synthetic nucleic acids come only from providers that screen orders for dangerous sequences.

  20. Direct regulators to clarify how their regulations apply to AI agents.

    Direct regulators to clarify how their regulations apply to AI agents.

  21. Amend consumer law so AI developers are responsible for what their models can do, including risks their systems create, without liability shifting to businesses and consumers not well placed to address them.

    Amend consumer law so that AI developers are responsible for what their models can do, including the risks their systems create. Liability should not shift to Australian businesses and consumers when they are not well placed to address those risks.

  22. Consider an optional AI Training Permit for training in Australia; training without a permit must comply with existing laws, including copyright.

    Entities conducting AI training in Australia may apply for an AI Training Permit. Training conducted without a permit must comply with existing laws, including copyright law.

  23. Under that model, set Government-determined fees paid into consolidated revenue, potentially tiered by purpose and scale, with low/no fees for limited commercial value, education and academic research.

    Fee. Permit holders would pay a fee into consolidated revenue. The fee would be set by Government and could be tiered to reflect differences in purpose and scale. Low or no fee permits could be available for training of limited commercial value, education and academic research.

  24. Under that model, require a free, effective, practical opt-out for Australian rights holders, designed with rights holders, AI companies and researchers.

    Consent and control. Permit holders would be required to respect an effective, practical, and free opt-out mechanism for Australian rights holders. The design of the opt-out mechanism would be developed in consultation with rights holders, AI companies, and researchers.

  25. Under that model, require compliance with National Data Centre Expectations, including energy, water, emissions and community commitments.

    Data centre expectations. Permit holders would be required to comply with the National Data Centre Expectations, including commitments on energy, water, emissions, and community impacts.

  26. Under that model, permit justified additional national-interest conditions, such as Government notification of training runs and transparency/disclosure requirements for models created in Australia.

    Other national interest conditions. The permit could include additional conditions where justified, such as giving notice to the Australian Government of training runs, transparency and disclosure requirements around models created in Australia, and a prohibition on circumventing technological protection measures (TPMs) within the meaning of the Copyright Act 1968.

  27. Under that model, authorise copyright-free training only for compliant permit holders; retain the prohibition on circumvention and withhold privileges from non-permit holders or those failing conditions.

    Permit holders would be authorised to conduct AI training in Australia with the legal certainty that such training would not be subject to copyright restrictions (the separate prohibition on circumventing TPMs continues to apply). The privilege attaches only to permit holders in compliance with their obligations. It does not apply to entities that do not hold a permit or that fail to meet its conditions.

  28. Under that model, have Industry, Science and Resources directly assist permit holders with local, state, territory and federal regulatory requirements.

    The Department of Industry, Science and Resources would also provide permit holders with direct assistance in understanding and managing regulatory requirements across local, state and territory and federal jurisdictions.

  29. Under that model, establish a separate, Government-administered creator fund from consolidated revenue, drawing on the Public Lending Right Act.

    Government would separately establish and administer a fund, resourced from consolidated revenue, to support Australian creators. The Public Lending Right Act 1985 provides a tested model.

  30. Under that model, compensate income loss and enrich Australian culture; limit eligibility to Australian citizen/permanent-resident creators, publishers publishing in Australia and their respective rights holders.

    The fund would serve a dual purpose: compensating Australian creators for income loss, and supporting the enrichment of Australian culture. Eligibility would be limited to Australian citizens and permanent residents who are creators, publishers who publish in Australia and their respective rights holders.

  31. Under that model, develop the fund's detailed design in consultation with creators to meet their expectations.

    The fund's detailed design would be developed in consultation with creators to ensure it meets their expectations.

  32. Under that model, amend the Copyright Act to link permit privileges to the scheme and apply privileges to all content protected by Australian law.

    Any policy will require some linkage between the Copyright Act 1968 and a broader scheme. Under this proposal, the permit is the vehicle for that linkage. Enabling amendments to the Copyright Act 1968 would grant privileges to permit holders. Those privileges would apply to all content protected by Australian law.

  33. Under that model, calculate fund payments from a funded pool with a tapering formula to support a broad creator base.

    Distribution. As with the Public Lending Right Act 1985, payments would be calculated from a funded pool, with a tapering formula to ensure the fund supports a broad base of creators rather than concentrating at the top.

  34. Under that model, provide departmental staff support for fund administration; have the committee set operational rules, with Ministerial approval via legislative instrument.

    Administration. As with the Public Lending Right Act 1985, the Minister would appoint a committee comprising relevant stakeholders — creators, publishers, and other appropriate representatives — to administer the fund. The committee would be supported by appropriate staff within the responsible department. Operational rules, including the mechanism for identifying eligible material and calculating payments, would be set by the committee and approved by the Minister via legislative instrument.

  35. Under that model, determine Government contributions through budget processes and ongoing policy judgment.

    Size of the Government contribution. As with the Public Lending Right Act 1985, the Government would decide how much to pay into the fund. The amount is a matter for budget processes and ongoing policy judgment.

  36. Under that model, use technical mechanisms lowering regulatory burden while ensuring effective opt-outs across all types of works.

    Effective. The approach should account for technical mechanisms that lower regulatory burden (such as machine-readable signals like robots.txt), while ensuring Australian rights holders can effectively opt out across all types of works.

  37. Under that model, allow simple opt-out by individual work, category or entire portfolio; avoid legal-knowledge, fee or complex-identification barriers and unlimited manual tracking by AI companies.

    Practical for rights holders and AI companies. An opt-out that requires legal knowledge, fees, or complex identification of works will not deliver meaningful control. Rights holders should be able to opt out simply, at the level of individual works, categories of work, or all of their work. Equally, an opt-out that requires manual tracking of an unlimited number of works would not achieve the goal of facilitating economically viable AI training in Australia.

  38. Under that model, require reasonable efforts and demonstrable processes respecting opt-outs; impose escalating fines up to permit loss and give creators a practical complaints avenue.

    Enforceable. Permit holders must make reasonable efforts to respect opt-outs as a condition of their permit and be able to demonstrate the processes that they use to ensure compliance. Non-compliance should result in fines that escalate to the loss of the permit. Creators should have a practical avenue to raise concerns.

  39. Under that model, preserve technological-protection-measure rules: prohibit circumvention and make breaches grounds for scheme enforcement.

    Technological protection measures are a related signal of non-consent. Where a rights holder has applied a TPM to their material, the permit does not authorise circumvention. Permit holders must comply with Part V, Division 2A of the Copyright Act 1968, and breach is a ground for enforcement action under the scheme.

  40. Under that model, have Industry, Science and Resources enforce with the Attorney-General's Department through monitoring, investigation, financial penalties and permit suspension/withdrawal.

    Enforcement would be administered by the Department of Industry, Science and Resources (the department), which already administers the National Data Centre Expectations, in coordination with the Attorney-General’s Department. The department would have the power to monitor compliance, investigate breaches, and take action against permit holders, including imposing financial penalties and suspending or withdrawing the permit.

  41. Under that model, permit enforcement for unpaid fees, opt-out breaches, data-centre expectation breaches, unmet national-interest conditions or noncompliance with other Australian laws.

    Grounds for action by the department would include failure to pay the permit fee, failure to respect opt-outs, failure to comply with the National Data Centre Expectations, failure to meet other national interest conditions, or failure to comply with other Australian laws.

  42. Under that model, confine the scheme to domestic training; allow overseas-trained models in Australia without restricting deployment, use or adoption.

    The scheme regulates AI training in Australia only. It does not restrict the deployment of AI models in the Australian market. Models trained in other jurisdictions — including in jurisdictions with fair use or TDM exemptions — can continue to be offered to Australian users. The scheme is focused on attracting AI training activity to Australia, supporting Australian creators, and securing the benefits of AI. It does not cover AI use, deployment, or adoption.

  43. Under that model, preserve copyright rights for overseas training, training outside a permit, AI outputs and other uses.

    Creators retain their existing rights under the Copyright Act 1968. The AI Training Permit only authorises AI training conducted in Australia by permit holders in compliance with the scheme. It does not affect rights with respect to AI training occurring overseas, training not under a permit, AI outputs, or any other use of copyrighted material.

  44. Under that model, define Australian training by compute location, including training wholly or partly on chips physically in Australian jurisdiction.

    Training in Australia is based on the location of the AI training compute. It covers AI training occurring wholly or partly using chips physically within Australia’s jurisdiction.

  45. Under that model, keep permit fees and creator-fund expenditure independent, with both flowing through consolidated revenue and no direct link between them.

    The fund and the permit are decoupled. Permit fees flow into consolidated revenue. The fund draws from consolidated revenue. The scheme would be designed so that there is no direct link between the two.

  46. Under that model, consider institutional proxies for fund eligibility/payments, such as collecting-society membership or catalogue listings; alternatively or additionally, support the creative industry more broadly, subject to further consultation.

    The fund would need a workable proxy. Options include using existing institutional infrastructure — such as membership of recognised collecting societies like the Copyright Agency or APRA AMCOS, or listing in established catalogues like the ISBN system. Alternatively or additionally, the fund could support the creative industry more broadly, in recognition of the benefits of the creative industry to Australia as a whole. The appropriate mechanism is a matter for further consultation.

  47. Following consultation, mandate a PLR-style fund committee including creator and rights-holder representatives.

    Legislation to create the fund should mandate a committee like the PLR Committee. ○ The new Committee should include representatives from creator and rights holder groups to increase confidence that the fund would be well administered from their perspective.

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

  48. Following consultation, consider mandated distributions partly through rights-holder groups, declared collecting societies or statutory-licence organisations, while also making funds available to unrepresented creators.

    Legislation to create the fund could mandate that distributions be made in part through rights holder groups, declared collecting societies or organisations that administer statutory licences. ○ Funds should also be available to creators not represented by rights holder groups.

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

  49. Following consultation, consider fund-certainty mechanisms such as an endowment and investment mandate like the Disaster Ready Fund.

    Legislation to create the fund could include mechanisms to increase confidence that future governments would contribute. For instance, there could be a specific endowment and an investment mandate, like the Disaster Ready Fund.

  50. Following consultation, consider sunsetting the permit and fund after a period, such as five years, only if no permits are issued; avoid sunset if permits issue to preserve investor certainty.

    The Permit and the Fund could sunset after a given period (e.g., five years) in the event that no permits are ever issued. This would address rights holder concerns about the legacy of the scheme if it is never used. ○ If permits are issued, the legislation should not sunset to ensure certainty for companies that invest in Australia on the basis of the permit scheme.

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

  51. Obtain legal advice on whether the opt-out must be global or may be Australian only.

    Legal advice is required on whether the opt-out mechanism should be global or can be Australian only.

  52. Clarify that the proposed permit is annual and the fund ongoing.

    Some commentary suggested that the fund would be a once-off payment. ○ The permit is intended to be annual and the fund ongoing.

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

  53. Pursue a government-negotiated solution if policymakers agree direct negotiations are not on track.

    Policy-makers who agree that a directly-negotiated solution (Scenario 1) is not on track should pursue a Government-negotiated solution (Scenario 2).

Read the original

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

Open in the interactive guide