Require AI developers to disclose the material used to train their models
30 submissions. Each submission counts once for this shared ask. Counts do not measure importance or merit.
A shared ask groups requests for substantially the same action from three or more submissions when the group was created. Groups remain listed if withdrawals reduce their counts. AI models summarised recommendations, grouped them and checked the groups. The label summarises the requests; it is not a quote. A shared ask does not mean agreement on anything else.
Theme: Training data transparency
Recommendations from submissions
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. Extracts are chosen by an AI model. Submissions are listed in submission number order; recommendations follow their order on each submission’s page.
Human Technology Institute, University of Technology Sydney Submission 18
Include a training-data transparency requirement in the Australian Standards for AI, with public input through the upcoming public consultation on the Standards.
The Australian Government should include a training-data transparency requirement in the Australian Standards for AI, which are currently being developed. In developing this requirement, the Government should provide an opportunity for public input via its upcoming public consultation process on the Standards.
UNSW AI Institute Submission 19
Require AI developers to keep auditable records and disclose sufficient information about the Australian creative, cultural, media and publicly funded research material used to train their models.
Require AI developers to keep auditable records and disclose sufficient information about the Australian creative, cultural, media and publicly funded research material used to train their models.
Australian Writers’ Guild (AWG) and Authorship Collecting Society Australia Aotearoa (ACSAA) Submission 40
Require AI companies to meet specific transparency obligations on use of ICIP, including what ICIP has already been used to train their models.
AI companies must have specific transparency obligations around the use of ICIP, particularly around what ICIP has already been used to train their models.
APRA AMCOS Submission 56
Legislate a training data record-keeping and disclosure obligation for any AI model made available in Australia, in detail sufficient for a rightsholder to determine whether their works have been used.
Legislate a training data record-keeping and disclosure obligation for any AI model made available in Australia, in detail sufficient for a rightsholder to determine whether their works have been used.
Alliance for Journalists' Freedom Submission 88
Require AI firms to transparently log and itemise training information in enough detail that copyright holders can see if their work was used and whether they are eligible for compensation.
Require AI firms to transparently log and itemise training information. SUch logs should contain enough detail that copyright holders can see if their work has been used, and if they are eligible for compensation.
A program corrected obvious text-recognition errors in this quote for display.
Music Victoria Submission 132
Require AI developers to provide appropriate information about copyright works and datasets used in training so creators and rights holders can identify use and enforce their rights.
2. Require transparency. AI developers should be required to provide appropriate information about copyright works and datasets used in training so creators and rights holders can identify use of their work and enforce their rights.
Koup Music Submission 139
Introduce transparency and record-keeping obligations for commercial AI developers.
Introduce transparency and record-keeping obligations for commercial AI developers — so rights holders can determine whether, and how, their work has been used in training.
IFPI Submission 144
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.
Dr Samantha Pillay OAM Submission 153
Regulate AI companies where they are reachable, through their Australian offices, revenue and government contracts: require transparency about training data, testing in Australia before model updates reach Australians, and opt-in training on Australians' personal data.
Second, regulate where the companies are reachable: they have Australian offices, revenue and government contracts. Require transparency about training data, no model update reaching Australians before it is tested here, and training on Australians' personal data to be opt-in.
National Association for the Visual Arts Submission 189
Introduce mandatory transparency requirements for companies that compile, create and use AI training datasets, including the sources of material.
NAVA supports mandatory transparency requirements for companies that compile, create and use AI training datasets, including transparency about the sources from which material has been obtained.
Media, Entertainment, and Arts Alliance Submission 195
Introduce mandatory reporting of all materials used to train AI and all materials captured under AI licensing deals.
Recommendation 2: Introduce mandatory reporting of all materials used to train AI and all materials captured under AI licencing deals.
Local & Independent News Association (LINA) Submission 197
Require developers to disclose training use of copyright material in enough detail for publishers to identify their work, supervised and enforceable by a regulator.
AI developers must be required to disclose use of copyright material in training, at a level of detail sufficient for a publisher to determine whether its work is included. This obligation should be supervised and enforceable by a regulator.
A program corrected obvious text-recognition errors in this quote for display.
AQ Intelligence (AQI), Guardrail Technologies Australia (GTA) and First Nations Strategic Alliance (FNSA) Submission 202
Legislate the copyright settlement on a consent basis: Australian creative works not used to train AI without creator consent, with training-data transparency, workable licensing, and clarity on ownership, provenance and liability for AI outputs.
Legislate the copyright settlement on a consent basis. Give effect to the Government’s stated position that Australian creative works not be used to train AI without creator consent, backed by training-data transparency and workable licensing, with clarity on ownership, provenance and liability for AI outputs.
Australasian Music Publishers' Association Limited (AMPAL) Submission 209
Introduce proportionate, machine-readable, work-level transparency throughout model development, including third-party datasets and web-scraped material, and require preservation of copyright-management information.
introduce proportionate, machine-readable, work-level transparency throughout model development, including for material obtained through third-party datasets or collected by web- scraping tools, and require the preservation of copyright-management information so rightsholders can identify unauthorised uses and enforce their rights.
Forum on Information and Democracy Submission 252
Create a legal obligation for AI deployers to provide training dataset information for public scrutiny in an accessible form, including a searchable database.
Creating a legal obligation for AI deployers to provide information about their training datasets for public scrutiny in an easily accessible and understandable manner, including a searchable database.
Global Institute for Women's Leadership, ANU; Wiyi Yani U Thangani Institute; Victorian Women’s Trust; Professor Ramona Vijeyarasa; and Dr Leonora Risse Submission 254
Require transparency in training datasets used for AI development, including authorship, content and sex- and gender-disaggregated data.
Require transparency in training datasets used for AI development (e.g. attention to authorship, content, sex- and gender-disaggregated data).
ABC Submission 304
Require AI companies to be transparent about the data used to train their models and to provide real-time access to proprietary data they continue to ingest.
there need to be laws which require AI companies to be transparent about the data they have used in training their models and to provide real-time access to proprietary data that they continue to ingest.
A program corrected obvious text-recognition errors in this quote for display.
Australian Recording Industry Association (ARIA) and Phonographic Performance Company of Australia (PPCA) Submission 306
Introduce meaningful training-input record keeping and disclosure without onerous burdens on AI companies, such as ARIA’s two-tier approach.
Meaningful record-keeping and disclosure measures (such as the two-tiered approach proposed by ARIA in its submission to the Productivity Commission 24) would enable rightsholders to better manage and protect their works without imposing onerous burdens on AI companies.
Australian Lawyers Alliance (ALA) Submission 310
Ensure any AI copyright framework makes rights practically enforceable through proportionate transparency and record-keeping obligations letting rights holders ascertain use of protected material, identify the responsible entity and pursue a remedy.
any framework must ensure that the rights it confers are capable of practical enforcement. This requires proportionate transparency and record-keeping obligations sufficient to enable rights holders to ascertain whether protected material has been used, identify the responsible entity and pursue an effective remedy.
Australian Society of Authors (ASA) Submission 312
Impose transparency obligations on AI developers, via the Copyright Act or other legislation, so rightsholders can know if their works were in datasets used to train models deployed in Australia.
We request new transparency obligations be imposed on AI developers via the Copyright Act (or other legislation) so that Australian rightsholders can know if their works were included in datasets relied upon for AI training of models deployed in Australia.
Rightsholder Group Submission 329
Implement mandatory AI dataset and service transparency so rightsholders can identify use of their works, check licensing compliance and enforce their rights.
We encourage the implementation of mandatory transparency requirements on companies which compile, create and use AI training datasets and offer AI services to the general public. Among other things, to make AI companies accountable, rightsholders must be able to identify when their works have been accessed or used, that licensing conditions have been complied with, and to enforce their rights when necessary.
Mr Michael Sanderson Submission 363
Require commercial AI developers to provide sufficient transparency about protected Australian material for copyright to be enforceable.
The Committee should recommend that commercial AI developers provide sufficient transparency concerning protected Australian material for copyright rights to be enforceable
Mr Isaac Dugdale Submission 427
Require licensees to declare their training corpora so payment reaches the right people and artists can check whether their work was used.
Licensees must declare their training corpora, so money reaches the right people and artists can check whether their work was used.
Name withheld Submission 443
That disclosure of training data provenance be mandated for any AI model offered commercially in Australia, at a granularity sufficient for a rights holder to determine whether their work was used, with enforceable penalties.
Mr Colin Cassidy Submission 444
Condition major-platform market access on training-data transparency and records, a mandatory Australian representative for service of process, platform-funded independent audits and civil penalties scaled to global turnover.
Make compliance a condition of market access for major AI platforms, including training-data transparency and record-keeping, a mandatory Australian legal representative for service of process, platform-funded independent auditing, and civil penalties scaled to global turnover.
Mr David Hynes Submission 447
Establish regulator-mediated training-data transparency for commercial models offered in Australia, with rightsholder query rights and no forced trade-secret disclosure.
regulator-mediated training-data transparency for commercial models offered in Australia, with rightsholder query rights and no forced trade-secret disclosure;
Name withheld Submission 477
Require AI providers to register training datasets with a public registry (managed by IP Australia); creators can flag works for exclusion from training
Mr Sam Moran Submission 507
Require developers to document and disclose training use in ways creators can understand and act on, supported by audit and redress mechanisms.
Shifting this burden requires developers to document and disclose training use in ways creators can understand and act on, supported by audit and redress mechanisms.
Design Institute of Australia Submission 551
Introduce three levels of training-data transparency: a public summary of main content types and sources, detailed records for regulators, and a process for rights holders to ask whether a work was used.
Introduce three levels of training-data transparency: a public summary of the main types and sources of training content; detailed records available to regulators; and a clear process for designers and other rights holders to ask whether a particular work was used.
Mr Nathan Brenner Submission 565
Require developers to keep auditable records and produce them when a claim is raised, rather than leaving creators to reconstruct commercial systems.
Transparency has to carry legal force, with a duty on the developer to keep auditable records and produce them when a claim is raised, so a creator is not left to reverse-engineer a commercial system just to begin asking.