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 144

IFPI

Submitter group: Creative sector.

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.

The global recording industry body argues Australia should keep its authorisation-based copyright framework, reject a text-and-data mining exception and compulsory licensing, and let the AI licensing market develop. It recommends enforceable training-data disclosure obligations, proportionate AI labelling, protection against unauthorised digital replicas of artists (after assessing existing laws), and Australian law applying to AI supplied into Australia.

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 Australia should keep its existing copyright framework of exclusive rights and prior authorisation, which it says is already fit for purpose for AI.

    Australia should maintain its existing copyright framework based on exclusive rights and prior authorisation, which is already fit-for-purpose for AI and aligns closely with international copyright standards.

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

  2. Argues that Australia should keep rejecting a text-and-data mining exception, because it says an AI licensing market is already developing.

    IFPI recommends that Australia continues to reject proposals for a text-and-data mining (TDM) exception or other limitation permitting protected content to be used for AI development without consent. Such intervention would be unjustified given the rapidly developing AI licensing market.

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

  3. Argues that compulsory or statutory licensing, levies and mandatory collective licensing would lead to under-remuneration and less innovation.

    Compulsory or statutory licensing, levies, mandatory collective licensing or other compulsory remuneration mechanisms for AI development will lead to under-remuneration and less innovation in creative economies

  4. Proposes a two-tier approach giving right holders information to enforce rights while protecting AI developers against frivolous requests and competitors seeking commercially valuable information.

    This could be done via a two-tiered approach that can strike a balance between right holders and AI developers, ensuring legitimate parties have the necessary information they require to exercise and enforce their legitimate rights, while AI developers are protected from frivolous requests or attempts by competitors to acquire commercially valuable information.

  5. Recommends clear labelling of content in which generative AI played a material role, without labelling negligible uses.

    IFPI therefore supports the clear labelling of content in which generative AI has played a material role, taking account of voluntarily agreed industry standards. Negligible or non-material uses of AI should not trigger labelling requirements.

  6. Recommends that AI products, services and outputs supplied to the Australian market comply with Australian copyright law wherever trained or generated.

    the Australian Government should strive to create a level playing field for copyright holders and AI companies alike, by ensuring that all entities that supply AI products, services or outputs to the Australian market comply with Australian copyright laws regardless of where the system was trained or the output generated.

Recommendations

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

  1. Maintain the authorisation-based copyright framework and reject any new TDM or AI-training exception.

    maintain its authorisation-based copyright framework and reject any new TDM or AI-training exception;

  2. Continue to observe ongoing international developments before recommending any legislative proposal on copyright and AI at this stage.

    We therefore urge the Australian Government to continue to observe ongoing developments before recommending any legislative proposal at this stage.

  3. Reject compulsory or statutory licensing, levies, mandatory collective licensing and other compulsory remuneration mechanisms for AI development.

    The Australian Government should also reject any proposals to introduce compulsory or statutory licensing, levies, mandatory collective licensing or other compulsory remuneration mechanisms for AI development.

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

  4. Introduce meaningful, actionable and enforceable record-keeping and disclosure obligations for developers of generative AI models regarding the content used for training.

    we urge the Australian Government to introduce meaningful, actionable and enforceable record-keeping and disclosure obligations for developers of generative AI models regarding the content they used for training.

  5. Consider two-tier disclosure that gives legitimate right holders information to exercise and enforce rights while protecting developers against frivolous requests and competitors seeking commercially valuable information.

    In order to promote trust and the development of the already burgeoning AI licensing market, we urge the Australian Government to introduce meaningful, actionable and enforceable record-keeping and disclosure obligations for developers of generative AI models regarding the content they used for training. This could be done via a two-tiered approach that can strike a balance between right holders and AI developers, ensuring legitimate parties have the necessary information they require to exercise and enforce their legitimate rights, while AI developers are protected from frivolous requests or attempts by competitors to acquire commercially valuable information.

  6. Under proposed tier one, developers and deployers should publish training-data or protected-content information enabling right holders to determine prima facie whether and how their rights were engaged and act.

    Tier one: AI developers and deployers make publicly available sufficiently detailed publicly information of the use of training data or copyright-protected content to available enable right holders to make a prima facie determination if and how their summaries rights have been engaged and, subsequently, to take appropriate action; and

  7. Under proposed tier two, give requesting right holders enforcing or licensing rights full records of legitimate interest, including music identifiers where available, sources, location/date/time of acquisition and training use, and legal basis.

    Tier two: Upon request by right holders seeking to enforce or license their rights, more detailed provide access to full records of training data for which the party has a records upon legitimate interest, including copyright-protected content. request For recorded music, this would include: ▪ information to enable the identification of specific works, where this information was available at the point of access/use (for sound recordings, this could be the track title, recording artist, and/or ISRC (International Standard Recording Codes)). ▪ the source of the content (e.g., specific URLs/datasets used). ▪ the location, date, and time at which the content was obtained and used for training. ▪ the legal basis for its access and use.

  8. Reject developers’ trade-secret protections for content access infringing copyright and circumventing technical or contractual restrictions; weigh concerns against creators’ rights, protecting genuine business secrets where necessary.

    It is therefore critical (and justifiable) to assert that AI developers cannot avail themselves of trade secret protections where copyright-protected content has been accessed by infringing the rights of copyright holders and circumventing access restriction measures, general or specific for TDM, implemented by technical means or by contractual terms/in the Terms of Service. Moreover, the Australian Government should weigh these concerns against the fundamental and exclusive rights of Australian creators who fuel the very innovation AI seeks to enhance. The Australian creative sector is crucial to the country’s economic development; allowing it to be undermined by opaque AI practices actively harms the sector. Australian law contains existing procedures to protect genuine business secrets, where necessary.

  9. Support clear labelling of content in which generative AI has played a material role, taking account of voluntarily agreed industry standards, with negligible or non-material uses not triggering labelling.

    IFPI therefore supports the clear labelling of content in which generative AI has played a material role, taking account of voluntarily agreed industry standards. Negligible or non-material uses of AI should not trigger labelling requirements.

  10. Support compatibility with industry labelling approaches, with regulatory requirements flexible enough for evolving technology and market practices.

    These industry-led initiatives demonstrate that practical and proportionate approaches are being developed to help music consumers make informed choices, promote transparency around the use of AI in music, and ensure that innovation develops in a way that continues to recognise and celebrate human artistry. The Australian Government should support compatibility with such approaches and ensure that any regulatory requirements are sufficiently flexible to accommodate evolving technology and market practices. Responsibilities for labelling AI-generated and AI-assisted content should reflect the different roles of participants across the AI value chain.

  11. Allocate labelling roles across the AI value chain: generation providers apply or enable labels; downstream services preserve identifiers and inform users; do not place the whole burden on artists or right holders.

    Responsibilities for labelling AI-generated and AI-assisted content should reflect the different roles of participants across the AI value chain. Providers of systems that generate content will generally be best placed to apply or enable the application of reliable labels and machine-readable information at the point of generation. Platforms and other downstream services should preserve relevant identifiers and communicate appropriate information to users. The burden should not fall entirely on artists, record companies or other right holders that may not control or possess information about the AI system used.

  12. Set clear outcomes for labelling legislation rather than mandating technical standards that may become outdated.

    Labelling mechanisms may include both clear and conspicuous disclosures appropriate to the relevant medium and machine-readable information. Relevant actors should apply or maintain labels in accordance with their respective roles and capabilities, so that information applied at the point of generation is not lost as content moves through the supply chain. Any legislation should set clear outcomes rather than mandate a particular technical standard that may quickly become outdated.

  13. Ensure labelling does not replace authorisation or legitimise outputs derived from unauthorised uses of recordings or artists’ identities.

    Labelling must not replace authorisation or otherwise legitimise outputs derived from unauthorised uses. Identifying an output as AI-generated or as a digital replica does not authorise the use of a protected recording or an artist’s voice, image, name or likeness.

  14. Protect artists’ voice, image, name and likeness: assess existing laws, material gaps and enforceable remedies before considering legislation; preserve contractual freedom and established licensing and enforcement practices in any reform.

    The Australian Government should ensure that artists have effective protection against unauthorised uses of their voice, image, name and likeness. With this in mind and before considering new legislation, the Government should assess the protections available under existing laws, identify any material gaps and ensure that available remedies are practically enforceable. Any reform should preserve contractual freedom and established licensing and enforcement practices, so that artists remain free to authorise and commercialise legitimate uses of their identity.

  15. Apply existing copyright rules, including primary and secondary liability, consistently where AI developers, providers or users copy or make available protected works without authorisation.

    Existing copyright rules, including those concerning primary and secondary liability, should apply consistently where the activities of AI developers, providers or users involve the unauthorised copying or making available of protected works or sound recordings.

  16. Ensure all entities supplying AI products, services or outputs to the Australian market comply with Australian copyright laws regardless of where the system was trained or the output generated.

    ensuring that all entities that supply AI products, services or outputs to the Australian market comply with Australian copyright laws regardless of where the system was trained or the output generated.

  17. Make any additional AI regulation establish clear responsibility for compliance and provide proportionate but effective remedies and sanctions.

    Any additional AI regulation should establish clear responsibility for compliance and provide proportionate but effective remedies and sanctions.

Read the original

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

Open in the interactive guide