Consultation on AI and Copyright

Overview

Our responses to the consultation reflect the need to safeguard the voiceover industry and ensure ethical use of AI technologies. This paragraph is an overview of our approach, with sections below per question (omitting questions 1-4 as these are just contact details).

We proposed several measures to protect creators, particularly voiceover artists, whose work is deeply personal and vulnerable to misuse in the AI era.

Key points included:
1. Rejecting the Data Mining Exception (Option 3): We’ve opposed Option 3 and similar exceptions, as they impose undue burdens on creators to monitor and opt out. Instead, we advocate for Option 1, requiring explicit licensing in all cases to protect the unique identity of voiceover artists and prevent unauthorised cloning.

2. National Voice Database and Challenge Phrases: A centralised voice database would allow artists to register their voice likeness and require explicit consent for any cloning or use in training AI models. To ensure compliance, all cloning processes should require a challenge phrase recorded by the individual to confirm consent. This would prevent misuse and provide creators with control over their voices.

3. Transparency and Accountability: We support requiring AI developers to disclose their training data sources, maintain auditable documentation, and integrate machine-readable rights reservations. This would ensure companies respect creators’ rights and allow enforcement of copyright protections. Platforms like YouTube, social media, and broadcasters should also be required to provide mechanisms for reporting unauthorised use of voices or likenesses.

4. Restricting Synthetic Data and Feedback Loops: AI-generated content without human input should be barred from use in training new models. Synthetic data derived from non-commercial sources, such as research institutions, should also be prohibited from training commercial models. These measures prevent a self-perpetuating cycle of misuse and protect public research.

5. Tiered AI Labelling Requirements: We propose categorising AI use into three levels: AI minor use (e.g., basic audio clean-up, no labelling needed), AI enhanced (e.g., performance transformation, requiring watermarks and metadata), and AI generated (e.g., text-to-speech synthesis, requiring clear labelling and visible watermarks). This ensures transparency while avoiding over-regulation.

6. Reforming Computer-Generated Works (CGW): We advocate removing copyright protection for works created without substantial human input. All AI-generated works should be marked as such and prohibited from training new models by default. This protects creators and prevents exploitation.

7. Government Oversight and Licensing: AI companies should hold licences to operate in the UK, ensuring compliance with ethical and copyright standards. Fees from these licences could fund the development of labelling technologies and enforcement tools. The UK should also align its approach with international standards to close cross-border loopholes.

8. Enforcing Penalties for Non-Compliance: Strong penalties, including fines, criminal liability, and market exclusion, should apply to companies ignoring rights reservations or misusing copyrighted works.

Question 4

4. Do you agree that option 3 - a data mining exception which allows right holders to reserve their rights, supported by transparency measures - is most likely to meet the objectives set out above?

No. For voiceover artists and creatives, option 3 could mean constantly monitoring how their work is used and ensuring they’ve opted out where necessary. This creates an additional administrative burden.
Voiceover recordings represent not just creative output but the unique and identifiable personal likeness of an individual. Unlike artwork or music, a voice is intrinsic to a person’s identity and can be misused to clone or imitate them without consent, impacting both their livelihood and personal integrity. This deeply personal nature of the voice warrants stronger protections and specific safeguards to prevent exploitation.

Question 5

5. Which option do you prefer and why?

Option 1: Strengthen copyright requiring licensing in all cases

For industries like voiceover recordings, where protecting unique, identifiable creative contributions is critical, Option 3 may fall short. Option 1, requiring explicit licensing in all cases would seem more appropriate.
Voiceover recordings represent not just creative output but the unique and identifiable personal likeness of an individual. Unlike artwork or music, a voice is intrinsic to a person’s identity and can be misused to clone or imitate them without consent, impacting both their livelihood and personal integrity. This deeply personal nature of the voice warrants stronger protections and specific safeguards to prevent exploitation.

Question 6

6. Do you support the introduction of an exception along the lines outlined in section C of the consultation?

No.
Voiceover recordings represent not just creative output but the unique and identifiable personal likeness of an individual. Unlike artwork or music, a voice is intrinsic to a person’s identity and can be misused to clone or imitate them without consent, impacting both their livelihood and personal integrity.
Thus in the context of voiceover recordings, requiring explicit licensing in all cases would seem more appropriate.

Question 7

7. If so, what aspects do you consider to be the most important?

n/a

Question 8

8. If not, what other approach do you propose and how would that achieve the intended balance of objectives?

To protect voiceover artists from unauthorised cloning or misuse of their voices by AI companies like ElevenLabs, I propose the creation of a national database of voice likenesses. Professional voiceover artists and other voice-dependent professionals, such as actors and broadcasters, could voluntarily register their voice likeness in this database. AI companies would be required to consult the database and obtain explicit permission directly from the individual, not just the copyright holder, before cloning or using a recognisably similar voice for training or generation.

As an additional safeguard, voice cloning (including quick cloning) should require the speaker to record a unique challenge phrase to confirm active, explicit consent for the cloning process. This ensures full control remains with the individual.

The database would be managed by a neutral authority, using robust voice comparison tools to automate the detection of protected voices and prevent companies from bypassing protections or accessing raw voice data. This approach respects the personal and professional value of a voiceover artist’s work while fostering transparency, accountability, and ethical AI practices. It would ensure AI systems are trained responsibly, protecting the rights and livelihoods of creators.

Question 9

9. What influence, positive or negative, would the introduction of an exception along these lines have on you or your organisation? Please provide quantitative information where possible.

The introduction of a data mining exception as outlined in Section C could have significant negative consequences for voiceover artists and businesses. Voice recordings are not just intellectual property but a representation of personal identity. Allowing AI developers to use such material without explicit permission undermines the artist’s control over their voice and risks exploitation through unauthorised cloning or imitation.

For our organisation, which represents and collaborates with professional voiceover artists, this could lead to:
Loss of income and livelihoods: If AI-generated voices become widely accessible without proper licensing, clients may substitute real artists with synthetic alternatives, significantly reducing demand for professional services.

Erosion of trust: Artists may hesitate to collaborate with organisations unable to guarantee that their work is protected, impacting business relationships and growth.

Administrative burdens: Monitoring and opting out of the exception would require additional resources, which could be prohibitive for smaller operations like ours.

Question 10

10. What action should a developer take when a reservation has been applied to a copy of a work?

When a reservation has been applied to a copy of a work, developers should be required to:

Cease Use Immediately: Halt any data mining, processing, or use of the reserved work/voice likeness in their systems or datasets.

Notify the Right Holder: Inform the right holder (or the artist, in the case of voice likenesses) that the reservation has been recognised and confirm compliance.

Remove the Work from Databases: Permanently delete any instances of the reserved work/likeness from their training datasets, models, or storage systems to ensure it cannot be reused inadvertently.

Provide Proof of Compliance: Submit a record or report to an independent authority verifying that the reserved work has been removed and detailing the steps taken to ensure compliance.

Question 11

11. What should be the legal consequences if a reservation is ignored?

If a reservation is ignored, the legal consequences should include financial penalties, injunctions to halt further use, and exclusion from the UK market for non-compliant companies. The UK should also work with international partners to enforce copyright protections across borders and ensure mutual recognition of reservations. Deliberate violations could lead to criminal liability for responsible parties, and companies should be required to delete any infringing datasets or models and compensate the creator.
For non-UK companies, compliance with UK standards should be mandatory for offering services to UK users, ensuring robust protections for creators while addressing the challenges of cross-border enforcement.

Question 12

12. Do you agree that rights should be reserved in machine-readable formats? Where possible, please indicate what you anticipate the cost of introducing and/or complying with a rights reservation in machine-readable format would be.

Yes, rights should be reserved in machine-readable formats as this would enable companies to identify and respect reservations automatically, reducing the risk of misuse. For voiceover artists, this would integrate seamlessly with the proposed database, allowing their voice likenesses to be tagged with machine-readable protections that AI systems can easily recognise. While implementation might incur initial costs for companies, these would be offset by streamlined compliance and reduced legal risks. For artists, it ensures their rights are upheld efficiently without needing constant manual enforcement.

Question 13

13. Is there a need for greater standardisation of rights reservation protocols?

Yes, there is a clear need for greater standardisation of rights reservation protocols. Standardisation ensures that creators, such as voiceover artists, can reliably and consistently protect their work across different platforms and jurisdictions. For example, a unified protocol integrated into the proposed voice likeness database would make it easier for companies to verify and respect reservations, fostering transparency and reducing the risk of unauthorised use. Additionally, standardisation would simplify compliance for AI developers, creating a common framework that balances the needs of creators and technological innovation.

Question 14

14. How can compliance with standards be encouraged?

Compliance can be encouraged by mandating adherence through regulation, offering financial incentives for adoption, and involving creators and developers in the creation of practical standards. Transparency measures, such as public disclosure of compliance, can further promote accountability and trust.

Question 15

15. Should the government have a role in ensuring this and, if so, what should that be?

Yes, the government should play a role by setting and enforcing regulations, supporting the development of standardised protocols, and ensuring compliance through penalties and oversight. It should also provide resources, such as tools or funding, to help creators and companies adopt these standards effectively.

Question 16

16. Does current practice relating to the licensing of copyright works for AI training meet the needs of creators and performers?

No, current practices often fall short of meeting the needs of creators and performers, including voiceover artists. Licensing frameworks are frequently opaque, leaving creators unaware of how their work is used or whether they are adequately compensated. Additionally, there is insufficient enforcement to prevent unauthorised use, such as scraping or cloning, which undermines creators' rights and livelihoods. A more transparent, standardised, and enforceable system is needed to ensure fair treatment and compensation for creators in the AI training process.

Question 17

17. Where possible, please indicate the revenue/cost that you or your organisation receives/pays per year for this licensing under current practice.

As a voiceover business, we currently do not receive specific revenue from licensing voice recordings for AI training, nor do we pay for such licensing. However, the lack of transparency in current practices means we are unable to fully ascertain whether our artists’ voices are being used without authorisation, potentially leading to untracked revenue loss for them. This highlights the urgent need for a clear and enforceable licensing framework.

Question 18

18. Should measures be introduced to support licensing good practice?

Yes, measures should be introduced to support licensing good practice. Clear guidelines, transparency requirements, and standardised licensing agreements would ensure that creators, such as voiceover artists, are fairly compensated and retain control over their work. Additionally, tools for tracking and managing licences could simplify compliance for businesses while safeguarding the rights of creators.

Question 19

19. Should the government have a role in encouraging collective licensing and/or data aggregation services?

Yes

Question 20

20. If so, what role should it play?

The government should establish clear regulations to standardise collective licensing frameworks, provide funding or incentives for the creation of data aggregation platforms, and oversee their operation to ensure transparency and fairness. Additionally, it should act as a mediator between creators, rights organisations, and AI developers to facilitate collaboration and ensure that creators’ rights are prioritised.

Question 21

21. Are you aware of any individuals or bodies with specific licensing needs that should be taken into account?

Yes, voiceover artists and performers have unique licensing needs due to the personal and identifiable nature of their work. Organisations representing voice talent, such as unions or trade associations (Equity etc), should be consulted to ensure their rights and concerns are addressed. Additionally, businesses like voiceover studios, which facilitate collaborations between artists and clients, require clear frameworks to protect both creators and their content.

Question 22

22. Do you agree that AI developers should disclose the sources of their training material?

Yes, AI developers should disclose the sources of their training material. This transparency is essential to ensure that creators, such as voiceover artists, can verify whether their work has been used without permission. Disclosure would also promote accountability, enabling rights holders to enforce their reservations and ensuring that training practices align with copyright laws.

Question 23

23. If so, what level of granularity is sufficient and necessary for AI firms when providing transparency over the inputs to generative models?

AI firms should provide transparency at a detailed level, including the following:

Specific Sources: Identify the datasets, recordings, or works used, including titles, creators, and licensing status.

Usage Context: Specify how the material was used in training, such as whether it was included in full, partially, or altered.

Permissions / clearance: Indicate whether permissions or licences were obtained, and from whom, for using each input. Provide evidence to show they regularly consult with a voice database likeness tool to check their training data is cleared for use.

This level of granularity ensures creators can verify and challenge the use of their work while enabling effective enforcement of copyright protections.

Question 24

24. What transparency should be required in relation to web crawlers?

Web crawlers should clearly disclose their purpose, the data they are collecting, and how it will be used, especially if it includes copyrighted material. They should identify the operating entity, respect opt-out mechanisms like robots.txt, and provide contact details. Transparency ensures accountability and protects rights holders from unauthorised data use. If no suitable explicit 'opt-in' is discovered then the web crawler should assume that the content is not suitable for use as AI training data.
Web crawlers who are storing audio for future voice training data sets should discard any audio files which include reserved voice likenesses.

Question 25

25. What is a proportionate approach to ensuring appropriate transparency?

A proportionate approach involves requiring AI developers and web crawlers to disclose their data sources and ensure compliance by cross-referencing a centralised voice database. This database would store creators’ rights and reservations in a machine-readable format, allowing developers to verify the suitability of training data or synthetic generation while respecting reservations. Standardised reporting frameworks and opt-out mechanisms, managed by a neutral authority, would balance transparency with minimal administrative burden for creators and companies.

Question 26

26. Where possible, please indicate what you anticipate the costs of introducing transparency measures on AI developers would be.

The costs for AI developers would depend on the complexity of their systems and the scale of their operations. Implementing transparency measures, such as integrating with a centralised voice database or reporting frameworks, might incur initial setup costs in the range of thousands to tens of thousands of pounds for system upgrades and compliance tools. Ongoing costs could include licensing fees for database access and administrative expenses for maintaining compliance, likely amounting to a few thousand pounds annually. While these costs could be significant for smaller developers, they are necessary to ensure accountability and fair treatment of creators.

Question 27

27. How can compliance with transparency requirements be encouraged, and does this require regulatory underpinning?

Compliance can be encouraged through regulatory underpinning, requiring AI developers to adhere to transparency standards as a legal obligation. Penalties for non-compliance, incentives for early adopters, and public disclosure of adherence to transparency measures would further promote accountability. A centralised system, such as the proposed voice database, could streamline compliance while ensuring consistent enforcement. Regulatory oversight is essential to ensure these measures are uniformly applied and effective.

Question 28

28. What are your views on the EU’s approach to transparency?

The EU’s approach, such as the AI Act, sets strong transparency standards by requiring AI developers to disclose training data and methodologies. While it promotes accountability, effective enforcement and cross-border applicability remain challenges. The UK could adopt similar principles while adding tailored safeguards, like a centralised voice database, to better protect creators such as voiceover artists.

Question 29

29. What steps can the government take to encourage AI developers to train their models in the UK and in accordance with UK law to ensure that the rights of right holders are respected?

While it may be difficult to incentivise AI developers to train their models in the UK, the government can encourage compliance with UK law by requiring developers to hold a licence to operate within the UK market. This licence would ensure their models meet UK standards for respecting the rights of creators and right holders. Access to the UK market, as an important and lucrative one, would act as a strong incentive, especially if UK laws align with international frameworks to minimise additional burdens. Fair and balanced regulations would further encourage developers to comply while maintaining accountability.

Question 30

30. To what extent does the copyright status of AI models trained outside the UK require clarification to ensure fairness for AI developers and right holders?

The copyright status of AI models trained outside the UK requires significant clarification to ensure fairness. Without clear guidance, there is a risk that models trained in jurisdictions with weaker protections could undermine UK creators’ rights when those models are used within the UK. Explicit rules should define how such models interact with UK copyright law, including whether their outputs can legally be used or commercialised here. This would create a level playing field, protecting right holders while giving AI developers certainty when operating in the UK market.

Question 31

31. Does the temporary copies exception require clarification in relation to AI training?

Yes, the temporary copies exception requires clarification in the context of AI training. This exception, originally intended for incidental and technical purposes, is increasingly being interpreted in ways that may allow data scraping and copying for AI training without proper licensing. Clear guidelines are needed to ensure the exception is not misused to bypass copyright protections, particularly for works like voice recordings, where unauthorised use can harm creators’ rights and livelihoods.

Question 32

32. If so, how could this be done in a way that does not undermine the intended purpose of this exception?

The temporary copies exception could be clarified by explicitly limiting its application to truly incidental and technical processes, such as caching or buffering, while excluding deliberate copying for AI training purposes. This could be achieved by introducing clear definitions and examples in the legislation, distinguishing between technical necessities and creative use. Additionally, requiring AI developers to obtain explicit permission or licences for any non-incidental use of copyrighted material ensures the exception fulfils its intended purpose without undermining creators’ rights.

Question 33

33. Does the existing data mining exception for non-commercial research remain fit for purpose?

The existing data mining exception for non-commercial research needs clarification to ensure it is not exploited. While it supports academic progress, the UK should explicitly state that outputs and content generated from research models under this exception cannot be used as inputs to commercial models. This would close a significant loophole and prevent commercial entities from indirectly benefiting from non-commercial research without proper licensing, ensuring creators’ rights are respected and protected.

Question 34

34. Should copyright rules relating to AI consider factors such as the purpose of an AI model, or the size of an AI firm?

Yes, copyright rules should consider the purpose of an AI model and the size of the firm, but with safeguards to prevent exploitation. The definition of “small firms” must be clear and tied to strict criteria, such as turnover and independence, to avoid larger companies creating shell entities to bypass rules. Models benefiting from leniency, such as those created by small firms or for non-commercial purposes, should have an expiry date in return for their exemptions, requiring a re-evaluation of their compliance if they are sold, acquired, or scaled for commercial use. This ensures fairness and prevents loopholes that undermine the intent of copyright protections.

Question 35

35. Are you in favour of maintaining current protection for computer-generated works? If yes, please explain whether and how you currently rely on this provision.

No, I am not in favour of maintaining the current protection for computer-generated works. The existing rules, which grant copyright to computer-generated works with no human authorship, are outdated in the context of advanced AI. In the voiceover and creative industries, AI-generated outputs are often based on the works of creators, such as voiceover recordings, artwork, or scripts, without adequate recognition or compensation for the original creators. Protecting computer-generated works without human authorship risks undermining human creators’ rights and reducing accountability for how AI systems use source material.

Instead, I would advocate for revising this protection to exclude purely computer-generated works and focus on safeguarding the rights of human contributors whose work forms the basis for AI outputs.

Question 36

36. Do you have views on how the provision should be interpreted?

The provision for computer-generated works should be interpreted to exclude AI-generated outputs unless there is clear, substantial human authorship guiding the creation. Importantly, copyright protection for AI-generated works must not override the need for AI developers to obtain explicit permission from creators, such as voiceover artists, before using their work for training or output generation. This ensures that creators retain control over their data and are not relegated to after-the-fact compensation for unauthorised use. Any copyright protections for AI-generated works must prioritise safeguarding the rights of human contributors from the outset.

Question 37

37. Would CGW legislation benefit from greater legal clarity, for example to clarify the originality requirement? If so, how should it be clarified?

Yes, CGW legislation would benefit from greater legal clarity, particularly regarding the originality requirement. Without clear rules, copyright could be granted to AI-generated works created using cloned voices, allowing those outputs to be used for further training, compounding the misuse of the original creator’s voice. The law should explicitly require that CGW protection is only granted when there is clear human creative input and when all training data has been used with explicit permission. This would protect voiceover artists from exploitation and ensure AI outputs do not undermine the value of human creativity.

Question 38

38. Should other changes be made to the scope of CGW protection?

Yes, the scope of CGW protection should be restricted to prevent the misuse of outputs derived from cloned voices. Granting copyright to such works risks enabling those outputs to be used in further training, perpetuating the unauthorised use of voiceover artists’ recordings. To address this, CGW protection should only apply to works with substantial human input and where all training data is explicitly licensed. This change would ensure the voiceover industry is not harmed by AI-generated works that exploit human talent without proper consent.

Question 39

39. Would reforming the CGW provision have an impact on you or your organisation? If so, how? Please provide quantitative information where possible.

Significant positive impact

Yes, reforming the CGW provision would significantly impact the voiceover industry. If copyright were granted to CGW created using cloned voices, it would create a dangerous feedback loop where these outputs could be used for further training without the original creator’s consent. This would increase the risk of unauthorised voice cloning and reduce demand for professional voiceover artists. Reforming CGW provisions to require explicit licensing of training data and prevent copyright on works with unauthorised inputs would protect the industry and ensure creators maintain control over their voices.

Question 40

40. Are you in favour of removing copyright protection for computer-generated works without a human author?

Yes, I am in favour of removing copyright protection for computer-generated works without substantial human involvement. Copyright should only apply to works with meaningful human creative input. However, to prevent misuse, all AI-generated work without substantial human involvement should be clearly marked as such and, by default, barred from use in training other models. This ensures that AI-generated content does not perpetuate a feedback loop where potentially unethical training data, such as cloned voices, is repeatedly exploited. This approach prioritises ethical practices, protects creators like voiceover artists, and ensures accountability in AI development.

Question 41

41. What would be the economic impact of doing this? Please provide quantitative information where possible.

Removing copyright protection for computer-generated works without substantial human involvement would positively impact industries reliant on human creativity, such as voiceover and performance arts. It would deter misuse of voiceover recordings for training AI models, protecting creators from revenue loss caused by unauthorised cloning or replacement by AI-generated voices.
Additionally, businesses in the creative sector would avoid legal disputes and revenue losses resulting from the ambiguity of granting copyright to AI-only works. This reform would uphold economic stability for human creators while promoting ethical AI practices.

Question 42

42. Would the removal of the current CGW provision affect you or your organisation? Please provide quantitative information where possible

Minor positive effect

The removal of the current CGW provision would positively impact our organisation and the voiceover industry as a whole. Without copyright for computer-generated works, AI-generated content would not unfairly compete with professional voiceover artists, preserving demand for authentic human voices. Additionally, it would discourage the unauthorised use of voice recordings for training AI models, ensuring creators retain control over their work. Removing the provision aligns with our goal of protecting the livelihoods and creative contributions of voiceover artists.

Question 43

43. Does the current approach to liability in AI-generated outputs allow effective enforcement of copyright?

No, the current approach to liability in AI-generated outputs does not allow for effective enforcement of copyright. When AI models produce outputs based on training data, it is often unclear who holds responsibility for ensuring compliance with copyright law—the AI developer, the user of the model, or both. This ambiguity makes it difficult for creators, such as voiceover artists, to identify and hold accountable the parties responsible for unauthorised use of their work.

To ensure effective enforcement, liability must be clarified, with AI developers held accountable for how their models are trained and users responsible for the legal use of generated outputs. Additionally, mandatory documentation of training datasets and licensing agreements would provide transparency and enable creators to trace potential infringements and enforce their rights. This would close existing enforcement gaps and better protect creators in the voiceover industry and beyond.

Question 44

44. What steps should AI providers take to avoid copyright infringing outputs?

From an AI voice perspective, AI providers should take the following steps to avoid unauthorised cloning and copyright-infringing outputs:

1. Integrate with a National Voice Database: Providers must verify all training data and outputs against a centralised voice database where voiceover artists can register their voice likeness and reserve their rights. Explicit consent from the registered artist, not just the copyright holder, must be required for any cloning.

2. Require a Challenge Phrase for Cloning: Any voice cloning process, including quick cloning, should require the speaker to record a unique challenge phrase confirming their consent. This ensures that cloning can only occur with active, verified permission from the individual.

3. Mark and Track AI-Generated Voices: AI-generated voices should include embedded watermarks or metadata that identify them as synthetic, along with information about the model used and licensing restrictions. A public tool should also be provided for users to verify whether a voice is AI-generated, fostering transparency and compliance.

4. Filter Outputs Against Registered Voices: Providers must use robust filters to compare outputs to voices registered in the database, blocking any unauthorised cloning or replication of protected voices.

5. Maintain Transparent Documentation: Providers should document all voice data used in training, including sources, permissions, and licensing agreements, and make this information auditable to ensure accountability.

6. Adopt Proactive Monitoring: Regular audits of training datasets and outputs should be conducted to identify and address any unauthorised uses of protected voices.

By implementing these measures, AI providers can prevent copyright infringement, protect voiceover artists from unauthorised cloning, and maintain ethical practices in AI voice technology development.

Question 45

45. Do you agree that generative AI outputs should be labelled as AI generated? If so, what is a proportionate approach, and is regulation required?

Yes, generative AI outputs should be labelled, but the approach must recognise varying levels of AI involvement. In the audio world, three distinct levels of AI use should be defined:
1. AI Minor Use: AI is used minimally to enhance existing work (e.g., cleaning up audio or reducing noise). Labelling is not required, as this does not significantly alter the human input.
2. AI Enhanced: AI is used to modify or transform a human performance, such as mapping a performance to a new voice likeness. Outputs should include subtle watermarks and metadata, with labelling available for those who need to verify AI involvement.
3. AI Generated: AI is used to create content from scratch, such as text-to-speech synthesis or fully synthetic voice creation. These outputs should be clearly labelled as AI-generated, with visible watermarks and metadata to ensure transparency and accountability.

This tiered approach provides clarity, aligns with industry practices, and avoids over-labelling. Regulation is required to ensure consistency, particularly for AI enhanced and AI generated content, standardising labelling and watermarking while protecting creators like voiceover artists from unauthorised use or misrepresentation.

Question 46

46. How can government support development of emerging tools and standards, reflecting the technical challenges associated with labelling tools?

The government can support the development of tools and standards by requiring AI companies operating in the UK to hold licences, with fees directed toward funding research and development of labelling technologies. This approach ensures that those profiting from AI contribute to addressing its technical challenges. The government could also establish partnerships with industry and academia to create robust watermarking, metadata, and verification tools, ensuring outputs can be easily identified and traced. Additionally, funding could support open-source solutions to make these tools widely accessible and interoperable, fostering transparency and compliance across the AI industry.

Question 47

47. What are your views on the EU's approach to AI output labelling?

The EU’s approach, as outlined in the AI Act, focuses on transparency by requiring AI-generated outputs to be labelled, which is a positive step toward accountability. However, its broad application could lack nuance, as it doesn’t appear to differentiate between varying levels of AI involvement, such as minor enhancements versus fully AI-generated content.

A more effective approach would define levels of AI involvement, such as AI minor use, AI enhanced, and AI generated, and apply labelling requirements proportionately. While the EU’s framework encourages transparency, clearer distinctions would ensure it remains practical for industries like voiceover, where AI might be used for clean-up or for creating entirely synthetic voices. The UK could improve upon the EU’s approach by incorporating such tiered standards and aligning global practices for consistency.

Question 48

48. To what extent would the approach(es) outlined in the first part of this consultation, in relation to transparency and text and data mining, provide individuals with sufficient control over the use of their image and voice in AI outputs?

The approaches outlined in the consultation make progress on transparency but fall short of giving individuals, such as voiceover artists, sufficient control over the use of their voice and image in AI outputs. Transparency alone does not prevent unauthorised use or cloning, and stronger safeguards are necessary.

To ensure meaningful control, the following measures should be implemented:

• A Centralised Voice Database: Individuals should be able to register their voice likeness, allowing AI developers to verify consent before using a voice in training or generating outputs.
• Explicit Consent Mechanisms: Training datasets must include only voices or images with documented, explicit permission from the individual, not just the copyright holder.
• Challenge Phrase Requirement: For all types of voice cloning (including quick cloning), individuals should be required to record a unique challenge phrase, confirming their consent to the cloning process.
• Labelling Requirements: All AI-generated voices should be clearly labelled and watermarked, enabling creators and consumers to identify unauthorised uses.

Without these measures, individuals remain at risk of unauthorised cloning, exploitation, and economic harm, undermining their ability to control how their voice and likeness are used.

Question 49

49. Could you share your experience or evidence of AI and digital replicas to date?

Yes, we’ve encountered several issues related to AI and digital replicas. Some of our voiceover artists have been cloned without their permission, with their voices used in radio adverts. This case was referred to Equity, but current legislation provides inadequate protection, and there are no repercussions for the services used to perform the cloning.

Online forums, such as Discord communities for voice cloning services like ElevenLabs, reveal users openly sharing how they have successfully cloned actors and characters without permission, with some believing this is legally acceptable.

Additionally, larger companies have begun pushing aggressive agreements, effectively coercing voiceover artists into relinquishing their moral and usage rights, likely in anticipation of using AI in the future.

We’ve also observed a decline in work, with 70% of our artists reporting reduced revenue over the past 12 months, some of which can be directly linked to the rise of AI-generated voices replacing traditional voiceover work. These examples highlight the urgent need for stronger legislation to protect creators in the face of rapidly advancing AI technologies.

Question 50

50. Is the legal framework that applies to AI products that interact with copyright works at the point of inference clear? If it is not, what could the government do to make it clearer?

No, the legal framework is not clear. When AI systems generate outputs that interact with copyrighted works, such as using a voiceover artist’s voice to create a clone, it is often ambiguous whether this constitutes infringement. This makes it difficult for creators to protect their rights.

To improve clarity, the government should: 1. Define Legal Use at Inference: Clearly outline what constitutes lawful and unlawful use of copyrighted works at the inference stage. 2. Require Licensing or Consent: Mandate that outputs resembling copyrighted works, such as voice clones, require explicit licensing or permission from the creator. 3. Establish Accountability for Platforms: Broadcast and publishing platforms, such as YouTube, social media, TV, and radio, should be required to provide mechanisms for creators to report unauthorised use of their voice or likeness and take appropriate action.

These steps would help AI-generated outputs respect copyright and provide creators, such as voiceover artists, with tools to address unauthorised use of their work.

Question 51

51. What are the implications of the use of synthetic data to train AI models and how could this develop over time, and how should the government respond?

The use of synthetic data to train AI models has both benefits and risks. While it can reduce reliance on real-world copyrighted materials, it may still perpetuate infringement if the synthetic data is derived from unauthorised use of copyrighted works, such as cloned voices. Over time, synthetic data could create a feedback loop where AI-generated outputs are reused to train new models, further distancing accountability from original creators and undermining creative industries like voiceover.

The government should respond by:
1. Regulating Synthetic Data Sources: Require transparency and documentation of synthetic data to ensure it is not derived from unauthorised or copyrighted works.
2. Prohibiting Use of Certain Synthetic Data for Training: Ban the use of synthetic data created by research institutions, universities, or non-commercial models for training commercial AI models, preventing the loophole mentioned earlier.
3. Restricting Training on Synthetic Outputs: Prevent synthetic outputs without substantial human input from being used to train future models, avoiding a self-reinforcing cycle of potentially unethical practices.
4. Enforcing Oversight: Establish monitoring systems to ensure synthetic data use aligns with ethical and legal standards.

These measures would help to ensure synthetic data is used responsibly, protect creators like voiceover artists, and prevent commercial exploitation of publicly developed models.