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 285

Business Software Alliance (BSA)

Submitter group: Industry bodies.

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 digital sovereignty requires capability, collaboration, resilience and trust with access to global technology and data. Opposes origin-based procurement exclusions and data localisation. Annexes proposals on risk- and role-based governance, skills, infrastructure, privacy, content authenticity and copyright, including a lawful-access training exception and a pause on economy-wide AI-specific rules pending gap analyses.

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 digital sovereignty should be understood as an ecosystem built on four pillars: capability, collaboration, resilience and trust.

    Digital sovereignty should be understood as the ability to build and sustain a digital ecosystem based on four pillars: 1) capability; 2) collaboration; 3) resilience; and 4) trust.

  2. Argues that achieving sovereignty objectives depends on open, interoperable digital governance, and that cross-border data and investment flows support jobs, innovation and public welfare.

    Achieving these objectives depends on open and interoperable digital governance policies. Cross-border data and investment flows support jobs, innovation, and public welfare across sectors.

  3. Says governments should avoid blanket origin-based exclusions and prescriptive procurement mandates, which it argues undermine competition and limit user choice.

    Governments should avoid blanket, origin-based exclusions, including prescriptive procurement mandates that undermine competition, reduce technological capacity, and limit user choice.

  4. Argues that AI obligations should be applied according to each entity's role, since actors have different information and control different decisions, rather than on a one-size-fits-all basis.

    Because each actor has access to different information and controls different decisions, obligations should be applied according to each entity's role rather than imposed on a one-size-fits-all basis.

  5. Says, in an annexed Productivity Commission filing, that it agrees the Government should pause consideration of economy-wide AI-specific regulation, including mandatory guardrails, until gap analyses are completed.

    Until these gap analyses are completed, the Government should pause consideration on economy-wide AI-specific regulations, including potential mandatory guardrails for AI in high-risk settings, to avoid regulatory duplication and imposing unnecessary compliance burdens.

  6. In annexed policy papers and PC filings, proposes a training exception for commercial and non-commercial AI using lawfully accessed material.

    The Productivity Commission should recommend that the Government introduce a clear statutory exception that applies to both commercial and non-commercial AI training conducted on lawfully accessed materials.

Recommendations

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

  1. Make sovereignty frameworks support provider and technology choice, continuity and contingency planning rather than concentrating risk in one provider, stack or infrastructure model.

    Rather than concentrating risk in a single provider, technology stack, or infrastructure model, sovereignty frameworks should support choice, continuity, and contingency planning; restrictive, origin-based mandates can undermine these objectives.

  2. Encourage investment in capability and innovation through worker upskilling, non-discriminatory R&D incentives, technology entrepreneurship support, high-performance computing and publicly accessible digital infrastructure.

    Encouraging investment in capability and innovation through worker upskilling, non- discriminatory R&D incentives, support for technology entrepreneurship, high- performance computing, and publicly accessible digital infrastructure;

  3. Institutionalise collaboration with governments and industry while supporting an open, pro-competitive digital ecosystem free from arbitrary, discriminatory, disguised or unnecessary investment, trade or data transfer restrictions.

    Institutionalizing collaboration with governments and industry while supporting an open, pro-competitive digital ecosystem free from arbitrary, discriminatory, disguised, or unnecessary investment, trade, or data transfer restrictions;

  4. Use outcome-based public procurement to preserve access to best-in-class technologies.

    Using outcome-based public procurement to preserve access to best-in-class technologies;

  5. Promote interoperability through voluntary, industry-driven standards for data, cybersecurity and supply-chain risk management.

    Promoting interoperability through voluntary, industry-driven standards for data, cybersecurity, and supply-chain risk management;

  6. Use encryption and privacy-enhancing technologies to protect sensitive data without unnecessary geographic restrictions.

    Using encryption and privacy-enhancing technologies to protect sensitive data without unnecessary geographic restrictions;

  7. Build trust through governance, transparency, legal safeguards and trading-partner early-warning and consultation mechanisms to manage cross-border service risks, including continuity-of-access commitments.

    Building trust through governance, transparency, legal safeguards, and early-warning and consultation mechanisms between trading partners to manage cross-border service risks, including continuity of access commitments.

  8. Avoid blanket, origin-based exclusions, including prescriptive procurement mandates that undermine competition and limit user choice.

    Governments should avoid blanket, origin-based exclusions, including prescriptive procurement mandates that undermine competition, reduce technological capacity, and limit user choice.

  9. In Annex A, pursue multistakeholder, interoperable risk-based governance, balanced value-chain responsibilities and common international AI terminology and taxonomy, building on EU–US Trade and Technology Council work.

    Countries should work together to promote multistakeholder dialogue and develop a shared vision for a risk-based policy approach for addressing common AI challenges and advancing norms around responsible AI governance (e.g., risk-based approach to regulation, balanced responsibilities along the AI value chain). Global partners should also agree on common AI terminology and taxonomy, including building on ongoing work in the EU-US Trade and Technology Council.

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

  10. In Annex A, require high-risk AI developers and deployers to conduct impact assessments and publicly affirm compliance; assess consequential eligibility decisions about housing, employment, credit, education, public accommodation, healthcare or insurance.

    BSA supports requirements for organizations that develop or deploy high-risk AI to conduct impact assessments and publicly affirm that they have complied with this practice. An AI system may be high-risk if it makes consequential decisions that determine an individual's eligibility for and result in the provision or denial of housing, employment, credit, education, access to physical places of public accommodation, healthcare, or insurance.

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

  11. In Annex A, incentivise high-risk AI safety and security testing, with technical standards developed through voluntary, market-driven, consensus-based processes.

    Testing high-risk AI systems. BSA encourages measures that incentivize safety and security. Robust testing and evaluation of high-risk AI systems for safety, security, accuracy, and fairness is critical and is prioritized in the NIST AI Risk Management Framework, which BSA supports. Existing technical standards for AI testing are nascent and should be developed consistent with longstanding voluntary, market-driven, and consensus-based approaches to standards development.

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

  12. In Annex A, adopt comprehensive consumer privacy law establishing strong individual rights, clear handling duties, robust security, legitimate business data use and government enforcement.

    Adopting comprehensive consumer privacy laws. BSA supports comprehensive consumer privacy laws that establish strong consumer rights in their personal data, impose clear requirements on companies that handle that consumer data, provide robust security, promote the use of data for legitimate business purposes, and are backed by robust government enforcement.

  13. In Annex A, provide opt-outs from consequential decisions made solely by AI without human interaction that have legal or similarly significant effects.

    Providing targeted opt-outs for profiling. Consumers should have the right to opt out of consequential, automated decisions that are made solely by AI without human interaction and that have a legal or similarly significant effect on individuals.

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

  14. In Annex A, have government agencies follow NIST AI Risk Management Framework practices, including for procurement.

    Implementing the NIST AI Risk Management Framework (RMF). The NIST AI RMF is a flexible framework that can help organizations govern, map, measure, and manage AI risks. Government agencies should follow the practices set forth by the NIST AI RMF, including for procurement purposes.

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

  15. In Annex A, procure from multiple cloud providers rather than placing all agency data in a single cloud infrastructure.

    Pursuing multi-cloud procurement. Government agencies should work with multiple cloud providers to leverage the breadth of innovation occurring across the cloud industry. Agencies should not put all their data in one cloud infrastructure, but rather leverage multiple cloud service providers’ compute, AI, and other technologies. BSA supports agencies using multi-cloud in cloud purchasing.

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

  16. In Annex A, have governments invest in AI cybersecurity solutions to match malicious actors already using AI.

    Deploying AI to meet today’s challenges with today’s solutions. Governments must invest in AI-driven cybersecurity solutions to bolster their defenses and keep pace with malicious actors who are already using AI to improve their exploits.

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

  17. In Annex A, prioritise commercial software and trustworthy embedded AI for government services and operations; ensure policies allow low-risk commercial AI adoption.

    Enabling use of commercial sector AI applications. AI applications are and will continue to be built into commercial software, including that procured by governments. Governments should continue to prioritize adopting commercial software and embrace trustworthy AI solutions contained within it to enhance citizen services and improve operations. This requires governments to ensure AI policies do not inadvertently prevent the government from adopting low-risk commercial AI applications.

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

  18. In Annex A, enable cyber defenders to use AI flexibly for attack-surface analysis, threat detection and security outcomes.

    Harnessing AI to improve cybersecurity risk management. Policymakers should ensure that cyber defenders can flexibly use AI to provide an accurate understanding of organizations’ attack surfaces and improve threat detection and security outcomes.

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

  19. In Annex A, narrowly tailor national-security measures so they do not unnecessarily interfere with routine business transactions.

    Using narrowly tailored measures to address national security risks. BSA supports narrowly tailored efforts to protect national security that do not unnecessarily interfere with companies’ ability to conduct routine business transactions.

  20. In Annex A, facilitate AI use in critical infrastructure; focus mitigation on risks of AI overriding human control and endangering health and safety.

    Leveraging AI to improve critical infrastructure. BSA recognizes that AI can contribute significantly to developing and improving critical infrastructure, such as transportation. BSA supports policies that facilitate the use of AI to enhance critical infrastructure. Efforts to mitigate risks to critical infrastructure should focus on instances where there is a risk that an AI system could override human control and endanger the health and safety of individuals.

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

  21. In Annex A, recognise open-source AI's role and support both open-source and proprietary development.

    Continuing the development of open source AI. AI policies should recognize the key role that open source plays in AI development. BSA encourages rules that support both open source and proprietary systems.

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

  22. In Annex A, make high-value non-sensitive government datasets open and available in machine-readable digital formats.

    Continuing efforts to make public government data sets open and available in machine-readable digital formats. Government-generated data is an important asset that can serve as a powerful engine for creating new jobs, promoting economic growth, and enabling innovation in AI-related technologies. Governments collect and generate vast quantities of data that offer unique insights into virtually every facet of the modern world. To enhance AI innovation, governments should continue to prioritize the release of high-value, non-sensitive government data.

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

  23. In Annex A, increase R&D funding to support innovation, high-paying jobs and economic competitiveness.

    Increasing investment in R&D. Increased funding for R&D is essential to sparking innovation, growing high-paying jobs, and ensuring economic competitiveness.

  24. In Annex A, cooperate internationally to identify and support cross-border R&D challenges.

    Encouraging R&D cooperation. Countries should work together to identify and support R&D challenges across borders.

  25. In Annex A, broaden STEM education opportunities and improve training to develop a diverse skilled workforce.

    Improving access and support for STEM education. Broadening educational opportunities, improving training programs, and ensuring the development of a diverse workforce is needed to help meet the demand for skilled STEM workers.

  26. In Annex A, have government and industry fund alternative AI career pathways without degree requirements, including apprenticeships, community-college partnerships, digital skills certifications, boot camps and public service opportunities.

    Expanding workforce training and alternative pathways. Industry and government should invest in programs to support creating alternative pathways to the full range of AI careers; this includes those that enable workers to develop high-demand technology skills without the need for a bachelor’s or graduate degree. Programs like apprenticeships, partnerships with community colleges, digital skills training and certifications, boot camps, and public service opportunities are all important gateways to helping new and mid-career workers develop in- demand skills.

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

  27. In Annex B, develop AI workforce training and skilling tailored to role, including office productivity, data science, IT and executive leadership.

    Develop workforce training and skilling initiatives to create an AI-ready workforce and respond to the future of work. The needs will vary and may overlap based on workforce role (e.g., office productivity, data scientist, IT team, and executive leadership).

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

  28. In Annex B, partner with the private sector to create AI skills curricula for educational institutions.

    Partner with the private sector to create AI skills curricula for educational institutions.

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

  29. In Annex B, create and use existing public-private partnerships for current and future workers' training and upskilling.

    Create and leverage existing public-private partnerships on workforce training and upskilling programs for current and future workers.

  30. In Annex B, encourage skills-based hiring in public and private sectors.

    Encourage skills-based hiring practices in both the public and private sectors.

  31. In Annex B, promote resilient AI, energy and connectivity infrastructure investment to expand AI use.

    Promote investment in AI, energy, and connectivity infrastructure to strengthen their resilience and enable increased use of AI tools.

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

  32. In Annex B, remove barriers to advanced cloud/software adoption, including cross-border data restrictions and forced source-code or algorithm disclosures.

    Remove restrictions that impede the adoption of state-of-the-art cloud services and software necessary to enable enterprise AI adoption, including cross-border data restrictions and forced source code or algorithm disclosures.

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

  33. In Annex B, expand infrastructure and data access through public-private partnerships.

    Expand access to infrastructure and data through public-private partnerships

  34. In Annex B, advance technical interoperability for seamless integration of data and AI tools across contexts.

    Encourage policies that advance technical interoperability enabling seamless integration of data and AI tools across different contexts.

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

  35. In Annex B, encourage workable internal AI governance with clear review and decision-making processes.

    Encourage organizations to establish workable internal AI governance measures and policies with clear internal review and decision-making processes.

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

  36. In Annex B, promote clear national privacy and cybersecurity frameworks accounting for different AI value-chain roles, stages, uses and risks.

    Promote clear national frameworks that advance privacy and cybersecurity to support AI adoption and that account for different roles, stages, uses, and risk profiles along the AI value chain.

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

  37. In Annex B, prioritise enterprise security, enabling defensive AI and risk-based approaches using existing standards and best practices.

    Prioritize high levels of enterprise security to maximize the benefits of AI adoption, including enabling the use of AI to enhance cybersecurity solutions and promoting risk-based approaches that leverage existing standards and best practices.

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

  38. In Annex B, harmonise AI policy within and across jurisdictions to facilitate market access and reduce fragmented or duplicative obligations.

    Strengthen harmonization on AI policies among and within jurisdictions to ensure compatible policies that facilitate market access and reduce duplicative or fragmented obligations.

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

  39. In Annex B, remove barriers to agency enterprise-AI adoption by digitising and modernising data formats and streamlining cloud/AI procurement.

    Remove barriers to government agencies’ adoption of enterprise AI, including digitizing and modernizing data formats to enable more productive uses and streamlining the procurement process for cloud and AI tools.

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

  40. In Annex C, make legislation and policy assign AI obligations according to companies' different roles.

    Like other contexts, any legislation or policy framework that creates obligations for companies that design and use AI systems should reflect these different roles and assign responsibilities accordingly.

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

  41. In Annex D, defer appropriately to copyright-holder and AI-developer efforts on technological tools expressing training preferences.

    Governments should give appropriate deference to ongoing efforts by copyright holders and AI developers to develop new technological solutions for copyright holders to express their preferences regarding AI training.

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

  42. In Annex D, retain full effective copyright remedies for infringing AI outputs, as for other outputs.

    Copyright holders should have full and etfective remedies when their rights are infringed. This principle applies equally to outputs generated using AI systems and outputs generated in other ways. Copyright remedies have been effective to deter infringement and should remain so.

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

  43. In Annex D, consider steps to protect artists against wrongful dissemination of unauthorised digital replicas.

    Governments should also consider whether there are steps they can take to better protect artists from the wrongful dissemination of unauthorized digital replicas.

  44. In Annex D, protect copyright in work where generative AI enhances human creativity.

    When generative AI is used to enhance human creativity, the resulting work should be protected by copyright.

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

  45. In Annex D, legally permit non-consumptive analysis, including of material drawn from copyrighted works.

    Adopt or maintain policies that ensure non- consumptive analysis of any material (including material that may come from part of a copyrighted work) is legally permitted.

  46. In Annex D, encourage voluntary controls allowing rights holders to opt out of AI training.

    Encourage development of voluntary controls that permit parties to “opt out” of AI training.

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

  47. In Annex D, permit AI training with personal data subject to appropriate safeguards.

    Ensure personal data can be used to train AI tools, subject to appropriate safeguards.

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

  48. In Annex D, avoid requiring consent for every personal-data use; expressly permit purposes including legitimate interests so research is not discouraged or training data skewed.

    Promote a balanced set of privacy guardrails—and avoid requiring consent for all processing. Privacy laws should expressly permit personal data to be processed for a range of purposes, including legitimate interests, to avoid discouraging socially beneficial research or skewing data used to train AI tools.

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

  49. In Annex D, recognise AI training as a legitimate personal-data processing purpose.

    Recognize AI training as a legitimate purpose for processing data.

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

  50. In Annex D, promote privacy by design and accountability for personal-data handling in AI activities.

    Promote privacy-by-design and accountability frameworks for companies handling personal data in AI-related activities.

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

  51. In Annex D, distinguish regulator, user and public disclosures, focusing on high-risk rather than low-risk AI uses.

    Distinguish between disclosures that are appropriately made to regulators, to users, or to the public. Focus on disclosures of high-risk uses of AI tools, rather than low-risk uses.

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

  52. In Annex D, exempt trade secrets from transparency disclosure obligations.

    Protect trade secrets by clearly recognizing that transparency obligations do not require a company to disclose trade secret information.

  53. In Annex D, require high-level consumer-useful disclosures rather than granular training-dataset detail.

    Focus disclosure requirements on high-level information that helps consumers understand AI tools, without requiring granular detail about datasets used for training.

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

  54. In Annex D, promote trusted cross-border data transfers, avoid data localisation and harmonise jurisdictions for scalable international AI development.

    Promote responsible and trusted cross-border data transfers. @ Avoid data localization requirements. @ Support harmonization across jurisdictions to enable scalable, cross-border AI development.

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

  55. In Annex E, support a range of content-authenticity technologies rather than mandating a single technique.

    Policymakers should not require use of a single technique but should ensure companies can use a range of technologies to mark content as AI-generated or not.

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

  56. In Annex E, encourage mature provenance methods while keeping policy flexible for different technical and operational uses.

    Rather, policies should encourage the use of technical methods that are mature enough for adoption today while remaining flexible enough to accommodate a range of technical and operational uses.

  57. In Annex E, for AI-generated text focus on people knowing when they interact with AI.

    To ensure consumers know about text-based AI generated content, we recommend focusing on requirements that help individuals know when they interact with an AI system.

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

  58. In Annex E, accommodate evolving provenance technology in legislation, using open standards such as C2PA.

    Ensuring laws remain fit over time. What constitutes state of the art in ensuring solutions for content provenance will evolve over time; policymakers should ensure that any legislative framework accommodates such developments. Embracing open standards like C2PA can help in this effort.

  59. In Annex E, account for different services' risks in deepfake policy, recognising limited user-safety and public-order risks in B2B software.

    Policymakers should also recognize the different risks involved in distinct types of services. For example, business-to-business software services pose limited risk to user safety and public order, given the size of their user base and the fact that they do not provide services directly to consumers, which may create fewer deepfake-related concerns.

  60. In Annex E, require direct consumer-interacting AI services to identify themselves unless AI interaction is obvious.

    When companies provide AI systems that interact directly yar) interact with AI systems. with consumers they should tell consumers they are interacting with an AI system, unless it is obvious.

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

  61. In Annex E, apply provenance/authentication duties to consumer-facing audio, visual and video content; exclude text and B2B settings, where companies can choose methods appropriate to use.

    Focus obligations to use content provenance and know if content is authentication tools on consumer-facing audio, visual, or AI-generated. video content. These requirements should not apply to text or in business-to-business settings. This is important because using content authenticity mechanisms doesn't make sense for text—like when a consumer uses AI tools to edit the text of an email or translate a document. In business-to-business settings, companies can create separate methods to address content authenticity, based on their specific uses.

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

  62. In Annex E, let AI-content identification requirements be met using leading global tools including C2PA, avoiding country-specific standards that limit widely accepted approaches.

    Ensure requirements to identify AI-created content can be global technologies to satisfied by leading global tools, including open standards recognize AI-created like the Coalition for Content Provenance and Authenticity content. (C2PA). A range of countries are considering regulations on content provenance; adopting country-specific standards can limit consumers’ ability to use widely accepted standards that are updated over time.

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

  63. In Annex E, generally place AI-content provenance obligations on application developers that generate the content.

    Obligations to apply content provenance information to Al- by about AI-generated generated content are usually best placed on developers of AI content. AI applications—since AI applications are used to generate content. Companies can use tools like machine-readable watermarks, digital fingerprints, and secure metadata to identify the origin of that content.

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

  64. In Annex E, prohibit removal of provenance information such as machine-readable watermarks and secure metadata, except for security concerns.

    Prohibit stripping content provenance information such as ° 4 provenance information machine-readable watermarks and secure metadata, absent i—) remains available to users. security concerns.

  65. In the annexed June 2025 PC filing, if mandatory guardrails proceed, identify a specific, defined set of high-risk use cases to provide legal certainty.

    If DISR moves forward with developing mandatory guardrails, we strongly encourage the Government to identify a specific and defined set of use cases that are considered high-risk, which will provide legal certainty for organisations as to which new safeguards should be adopted and implemented.

  66. In the annexed June 2025 PC filing, define high-risk AI developers as entities designing specifically high-risk systems or substantially modifying systems that are or become high-risk.

    Developer of High-Risk AI System: An entity that: (1) designs an AI system specifically intended to be used as a high-risk system; (2) substantially modifies a high-risk AI system, or (3) substantially modifies a non-high-risk AI system so that it becomes a high-risk AI system.

  67. In the annexed June 2025 PC filing, limit deployers to entities actually using high-risk AI systems to deliver products or services.

    The definition of deployer should be narrowed to focus on those that actually use a high-risk AI system in delivering a product or service.

  68. In the annexed June 2025 PC filing, let organisations choose relevant content-provenance solutions under a framework accommodating technological change.

    any governance framework must be designed to accommodate such developments and should provide scope for organisations to assess what is the most relevant solution for them when it comes to content authentication and provenance mechanisms.

  69. In the annexed September 2025 PC filing, endorse comprehensive gap analyses of AI risks and their coverage under existing regulatory frameworks.

    The Government should undertake a comprehensive set of “gap analyses” to understand the risks stemming from AI and whether these risks can be dealt with under existing regulatory frameworks.

  70. In the annexed September 2025 PC filing, endorse pausing economy-wide AI-specific regulation, including mandatory high-risk guardrails, until gap analyses are complete.

    Until these gap analyses are completed, the Government should pause consideration on economy-wide AI-specific regulations, including potential mandatory guardrails for AI in high-risk settings, to avoid regulatory duplication and imposing unnecessary compliance burdens.

  71. In the annexed September 2025 PC filing, endorse economy-wide AI rules only as a last resort when existing frameworks cannot be sufficiently adapted and technology-neutral regulation is infeasible.

    Economy-wide AI-specific regulations should only be considered as a last resort for the use cases of AI that meet two criteria, namely: a) where existing regulatory frameworks cannot be sufficiently adapted to handle the issue; and b) where technology-neutral regulations are not feasible.

  72. In the annexed September 2025 PC filing, endorse mandatory high-risk guardrails only where gap analyses find existing frameworks or new technology-neutral rules cannot adequately mitigate harm.

    Mandatory guardrails for AI in high-risk settings are only appropriate in circumstances where the gap analyses determined that existing regulatory frameworks or new technology- neutral regulations cannot adequately mitigate the risk of harm.

  73. In the annexed September 2025 PC filing, ask the Commission to focus gap analyses on specific high-risk use cases.

    the Productivity Commission should make clear that the focus of the gap analyses is to identify high-risk AI use cases. This will ensure that any subsequent regulatory efforts are targeted at a set of specific use cases.

  74. In that PC filing, assess risk by intended application rather than sector: target consequential eligibility decisions with material legal or similarly significant effects concerning housing, employment, credit, education, public accommodation, healthcare or insurance.

    Generally, AI governance efforts should focus on addressing high-risk AI use cases based on potential harm. This approach assesses risk based on the AI system’s application and intended use case, rather than the sector in which it is used or deployed. In this regard, BSA has encouraged policymakers to focus on AI systems specifically developed to make consequential decisions, which are determinations that have a material legal or similarly significant effect on a consumer’s eligibility for and result in the provision or denial of housing, employment, credit, education, access to physical places of public accommodation, healthcare, or insurance.

  75. In the annexed September 2025 PC filing, ask the Commission to recognise distinct AI supply-chain roles and have agencies assess responsibilities accordingly.

    The Productivity Commission should recognise these distinct roles and urge agencies to consider the differences among them when assessing appropriate responsibilities.

  76. In the annexed September 2025 PC filing, ask the Commission to recommend a lead agency or non-regulatory coordinating body to prevent fragmented and duplicative AI policy.

    The Productivity Commission should urge the Government to officially designate a lead agency or non-regulatory central coordinating body to oversee AI policy coordination, with a focus on preventing internal regulatory fragmentation and duplication.

  77. In the annexed September 2025 PC filing, have the coordinating body set policy direction and coordinate risk-based, proportionate approaches, while leaving regulatory drafting to agencies and sector regulators.

    Importantly, the coordinating body’s function is not to draft regulations on behalf of agencies or sectoral regulators – rather, it should be setting the overarching policy direction, working closely with various Government stakeholders to coordinate regulatory approaches, and ensuring that AI governance across portfolios remains risk-based and proportionate.

  78. In the annexed September 2025 PC filing, ask for a statutory exception for commercial and non-commercial AI training on lawfully accessed material.

    The Productivity Commission should recommend that the Government introduce a clear statutory exception that applies to both commercial and non-commercial AI training conducted on lawfully accessed materials.

  79. In the annexed September 2025 PC filing, permit private and non-Australian DAT Act accreditation applicants that demonstrate compliance with Australian security, privacy and data-handling requirements.

    The Productivity Commission should recommend expanding the DAT Act accreditation framework to allow applications from both private and non-Australian entities, provided they can demonstrate compliance with Australia’s security, privacy, and data-handling requirements.

  80. In the annexed September 2025 PC filing, ask government to explore and support privacy-enhancing technologies alongside data-access reforms.

    The Productivity Commission should recommend that the Government explore and support the adoption of privacy-enhancing technologies to supplement efforts to improve data access.

  81. In the annexed September 2025 PC filing, prioritise the controller–processor distinction in the next Privacy Act amendments.

    The Productivity Commission should urge the Government to prioritise implementing the controller-processor distinction in the next tranche of Privacy Act amendments.

  82. In the annexed PC filings, align controller and processor definitions with leading privacy laws.

    1. Adopt definitions of controllers and processors that align with other important privacy laws. Under the GDPR, for example, a data controller is defined as an entity that “alone, or jointly with others, determines the purposes and means of processing personal data”, whereas a data processor is defined as an entity that “processes personal data on behalf of the controller”. The language provides a clear distinction between entities that decide how and why to collect an individual’s personal data and entities that act on behalf of others to process personal information.

  83. In annexed PC filings, expressly recognise context-specific controller/processor status: apply each role's duties when a company acts in that role.

    2. Clearly recognise that determining whether an entity is acting as a controller or processor is fact-based and context specific. A single company may act as a controller for some of its products and services (i.e., consumer-facing products) and as a processor for others (i.e., business-to-business services). Such a company should be subject to obligations the Act places on controllers when it acts as a controller and subject to obligations the Act places on processors when it acts as a processor. We recommend expressly stating this in the Act.

  84. In that annexed PC filing, assign controllers consent and consumer-rights duties; require processors to safeguard data and process it only on the controller's behalf and instructions.

    3. Ensure that obligations for controllers and processors are appropriately tailored according to their roles. Controllers and processors should be subject to distinct obligations under the Privacy Act. Importantly, consumer-facing responsibilities, such as obtaining consent from individuals and responding to consumer rights requests, should be assigned to controllers rather than processors. This reflects the fact controllers are entities that decide how and why to collect a consumer’s personal information. Processors should be subject to other obligations, such as requirements to safeguard personal information and to process personal information only on behalf of the controller and pursuant to its instructions. This allocation of responsibilities is also consistent with global norms.

Read the original

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

Open in the interactive guide