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 92

Australian Association of Voice Actors (AAVA)

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.

Describes income and work losses among voice actors, citing its 2026 voice-actor survey and testimonials, and argues for mandatory time-of-broadcast labelling of AI-generated content overseen by ACMA and the ACCC, a non-transferable biometric identity right, opt-in consent and Privacy Act expansion. Says existing bodies could oversee transparency but ACMA needs overhauled powers.

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. Says generative AI's impact has led 44.9% of voice actors to actively seek reskilling or work outside the creative arts.

    As a result of generative-AI’s impact on the voice acting industry, a staggering 44.9% of voice actors are now actively looking to re-skill or seek work outside of the creative arts.

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

  2. Argues that AI-generated content should be disclosed at the time of broadcast or exposure whenever the average consumer could believe it is real or human-produced.

    Disclosure at the time of broadcast, consumption of or exposure to AI-generate content is required when the average consumer could believe the content is real or human produced.

  3. Argues that self-regulation by the commercial radio industry does not work for AI transparency.

    Clearly self-regulation does not work, when an organisation such as CRA, whose members are also the media companies that profit from replacing its human workforce with AI-generated voices, gets to write its own rules.

  4. Argues that copyright law does not protect a performer's voice, likeness or biometric data, so payment for training works would not reach them.

    while copyright law protects creative works, it does not protect the voice, likeness, identity, or biometric data of the people whose performances are contained in those works. Paying for the IP or copyrighted works being used to train AI LLMs (such as AI-voice platforms) still fails to seek consent or pay compensation to these individuals.

  5. Says voice-cloning providers cannot or will not prove that their training data was obtained with consent.

    It should also be made clear that none of the TTS platforms can, or have willingness to, prove that their training data was obtained with consent.

  6. Argues that identifying the system that produced AI output will be crucial for scrutinising the training data behind it.

    the ability to identify the system which created a video, image, voice or text will be crucial in scrutinising the data (IP) which contributed to training of the AI system that generated the output.

Recommendations

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

  1. Have ACMA oversee guidelines for AI-generated content, with a specific focus on AI transparency.

    the Australian Communications and Media Authority (ACMA) would be the appropriate arm of government to oversee guidelines for the AI-generated content in the Australian media landscape, albeit with an additional, consistent and specific focus on AI transparency.

  2. Add mandatory at-time-of-broadcast labelling of AI-generated content to the broadcasting codes of practice that ACMA must register (commercial, community, subscription and open narrowcast radio and television).

    The following codes of practice, which currently must be registered by the ACMA, would be suited to include additions that cover mandatory at-time-of- broadcast labelling of AI-generated content:  Commercial radio code of practice  Community radio broadcasting codes of practice  Subscription narrowcast radio codes of practice  Open narrowcasting codes of practice  Commercial television industry code of practice  Subscription broadcast television codes of practice  Community television codes of practice  Subscription narrowcast television codes of practice  Open narrowcast television codes of practice

  3. Give ACMA a consistent AI transparency policy and the means to enforce it.

    A consistent and considered AI policy for ACMA is overdue, as is the means to enforce it.

  4. Give the ACCC the ability to require at-time-of-broadcast disclosure of AI-generated content in advertising, broadcast and online.

    Giving the ACCC the ability to require at-time-of-broadcast disclosure of AI-generated content in advertising (broadcast and online) would be a logical start.

  5. Consider having the ACCC require at-time-of-broadcast AI disclosure as an addition to the Australian Association of National Advertisers Code of Ethics.

    Giving the ACCC the ability to require at-time-of-broadcast disclosure of AI-generated content in advertising (broadcast and online) would be a logical start. The ACCC could require this as an addition to the Australia Association of National Advertisers (AANA) Code of Ethics

  6. Require disclosure of AI-generated content at the time of broadcast or exposure when the average consumer could believe it is real or human-produced.

    Disclosure at the time of broadcast, consumption of or exposure to AI-generate content is required when the average consumer could believe the content is real or human produced.

  7. Disclose AI-generated commercial content before broadcast and, where possible, during it; use both audible and visual labels where possible for accessibility.

    content produced with AI-generated content should be disclosed preceding, and where possible, during the commercial at the time of broadcast. For sight impaired people, audible labelling is essential for accessibility. For the hearing impaired, visual labelling is essential. This is why both should be employed where possible.

  8. Require AI disclosures to be clear, prominent, unavoidable and understandable.

    Unlike other advertising T&Cs or disclaimers (gambling or financial products) where an audible and/or visible disclaimer follows a commercial, content produced with AI- generated content should be disclosed preceding, and where possible, during the commercial at the time of broadcast. For sight impaired people, audible labelling is essential for accessibility. For the hearing impaired, visual labelling is essential. This is why both should be employed where possible. The Key Principle: Immediate Consumer Recognition. Disclosures Must be: • clear • prominent • unavoidable • understandable

  9. Apply disclosure requirements across television, radio, social media, digital display ads, influencer marketing, print advertising and advertorial journalism.

    Disclosures would apply across all media: • TV and radio • social media • digital display ads • influencer marketing • print advertising • advertorial journalism

  10. Prescribe Australian AI disclosure formats: visible image labels or icons, persistent video icons and opening notices, audible disclosures before and at appropriate intervals, and clear statements of AI-generated text origins.

    Format Labelling Images Visible label or icon. Video Persistent icon + opening notice. Audio Audible disclosure pre and at appropriate intervals. Text Clear statement of AI-generated origins. Following are practical demonstrations of how time-of-broadcast AI-labelling might look in Australia. Prescriptive formats will assure compliance and lessen the burden on regulators to assess if compliance has been reached.

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

  11. Consider visual labels and audible disclaimers before audiovisual commercials, and real-time visible supertitles where AI-generated images purport to be human.

    AUDIO-VISUAL COMMERCIAL (e.g. TV, Cinema, Online Video) A potential visual label with audible disclaimer preceding an audio-visual commercial, such as: or It is reasonable that AI-generated images purporting to be human are accompanied in real time by a visible supertitle, such as:

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

  12. Consider preceding visual labels for visual-only commercials and persistent onscreen logos where AI-generated images appear, including human representations.

    VISUAL ONLY COMMERCIALS (e.g. video without voice-over or dialogue) A potential visual label preceding a visual only commercial, such as: with a persistent logo onscreen throughout where AI-generated images appear. It would also be reasonable that AI-generated human representations be accompanied in real-time by a persistent visible logo, such as:

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

  13. Consider disclaimers before and after scripted cinema and television, without requiring the real-time disclosure applied to commercial content.

    A single disclaimer preceding a cinematic film or TV show, and then one following the work, would be a reasonable measure without expecting the same real-time disclosure during the episode or movie that commercial content would incur.

  14. Consider explicit AI-host identification through audible disclaimers before audio-only segments and audiovisual labels before audiovisual segments.

    When the job of a broadcaster is replaced by an AI-generated host it should not be assumed that the average listener or viewer would be aware of this unless they were explicitly informed. AUDIO ONLY PRESENTERS (e.g. Radio, Spotify, Podcast) A potential audible disclaimer preceding a segment in which the presenter/host is an AI-generated voice, could be: or AUDIO-VISUAL PRESENTERS (e.g. TV, Online) A potential audio-visual label preceding a segment where the presenter/host is Al- generated, such as:

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

  15. Consider real-time visible supertitles for AI-generated human-like audiovisual portrayals.

    It would also be reasonable that AI-generated human-like portrayals be accompanied in real time by a visible supertitle, such as:

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

  16. Consider clear, visible bylines identifying AI-generated articles, whether informative or promotional.

    WRITTEN ARTICLES & CONTENT Under the headline/title of an article (be it informative or promotional in nature) is usually the byline of its writer. In the case of AI-generated articles a clear and visible byline would be:

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

  17. Consider visible disclosure labels for AI-generated print and online images, whether informative or advertising.

    Whether printed and online content is created to be informative or for advertising purposes, reasonable disclosure should be expected. Al- generated images could have a visible label:

  18. Avoid AI-generated content and presenters in news and emergency information, or at least regulate them heavily.

    News and emergency information is an area in which AI-generated content and presenters should be avoided, or at the least heavily regulated.

  19. Consider audible disclosure for AI voices in news-programme recreations at the affected segment's start, plus a label and voiceover before the programme.

    In a recent CHANNEL 7 Spotlight story, Inside the mind of Erin Patterson, AI-generated voices were used in the dramatic re-creations of police interviews and the transcriptions of text messages. To their credit, Spotlight provided some, but minimal labelling; however, the program did not provide an audible disclaimer for the sight impaired. This could be verbally disclosed at the beginning of affected segment by the host or by voiceover, and a label and voiceover preceding the program.

  20. Have the independent regulator scrutinise the conflict of interest in commercial radio writing its own AI transparency rules.

    ACMA has shown no leadership by allowing these weak and insufficient measures to suffice as “transparency”, furthermore the conflict of interest that sees Commercial Radio write its own rules need further scrutiny by the independent regulator.

  21. Give ACMA a clear standard against which to measure adequate AI transparency.

    ACMA need a clear standard to measure adequate AI-transparency against, currently it does not have one.

  22. Mandate machine-readable watermarking of AI-generated content so the source and provenance of training data can be scrutinised.

    In discussions with The Tech Council of Australia, AAVA found that there were profound agreements on mandating machine-readable watermarking of AI-generated content so that the source and provenance of training data can be scrutinised easily. AAVA sees this as essential to protecting and enforcing proper use and consent.

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

  23. Introduce Biometric Identity Protections giving every Australian the legal right to control how or if their voice, image and likeness are captured, cloned and commercially used through AI, and be told when this occurs.

    AAVA is calling on the Australian Government to introduce Biometric Identity Protections for all Australians. Every Australian should have the legal right to control how/if their biometric data (voice, image, likeness, identity) are captured, cloned and commercially used through artificial intelligence; and be told when this has taken place.

  24. Consider legislation similar to the proposed US NO FAKES Act, alongside California's digital likeness protections.

    Australia should consider legislation similar to the United States' proposed NO FAKES Act34, alongside the multiple digital likeness protections recently enacted in California, to ensure Australians retain control over their own identity in the AI era.

  25. Make AI-cloning or deep-faking of deceased persons a highly restricted activity.

    AAVA recommends that the AI-cloning or “deep-faking” of deceased persons be a highly restricted activity.

  26. Consider data-protection legislation, a moral right and Privacy Act expansion for biometric data, prioritising a non-transferable, inalienable right as the protective foundation.

    Australia should consider a combination of legislative changes including a form of data-protection, a moral right, and an expansion of the Privacy Act (Cth) 1988* to include a person’s biometric data - but above all a non-transferrable inalienable right, such a moral right, would be the best foundation for protection.

  27. Provide a right to deletion when biometric data is used, and a right to be forgotten for deceased persons regarding use of their biometric data beyond death.

    Australians also deserve a “right to deletion” when their biometric data is used, anda “right to be forgotten” for diseased persons in relation to their biometric data being used beyond death.

  28. Overhaul ACMA's mandate, mission and scope of powers.

    an over-haul of ACMA’s mandate, mission, and scope of powers is well and truly overdue.

  29. Allow use of Australians' biometric data, especially monetised use, only on a strictly opt-in basis.

    The ONLY way an Australian citizens biometric data should be used (and especially monetised) should be on a strictly opt-in basis.

  30. Deem tech platforms that cannot show all biometric data used for training was used with full informed consent unfit to operate in Australia, with permission to operate only under a strict social contract.

    Tech Platforms (such as ElevenLabs and Fish.Audio) that do not offer full transparency and discoverability to scrutinise that all biometric data used to train their machines has been used with full informed consent should be deemed unfit to operate in Australia. The Australian Government must take a strong stance on consent: Permission to operate on our shores should happen only under a strict social contract.

  31. Impose very tight controls over where and to whom companies handling Australians' sensitive biometric and identity data send it.

    When a company is handling the sensitive biometric data and identity information of Australian citizens there should be very tight controls over where and to whom that data https://www.industry.gov.au/publications/australias-ai-ethics-principles is sent.

  32. Include personal data as well as biometric data in Privacy Act expansion.

    An expansion of the Privacy Act (Cth) 1988 48 to include a person’s biometric and personal data is essential.

  33. Expand moral rights laws to include a non-transferable, non-waivable, inalienable right giving Australians full control over use of their voice, likeness and image, including after death.

    An expansion of the Moral Rights laws for creators should be expanded to include a non-transferrable, non-waivable, inalienable right to give Australians full control over how their voice, likeness, image is used – even after death.

Read the original

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

Open in the interactive guide