What should the UK government do about copyright and AI?

The University of Oxford’s Institute for Ethics in AI published its report (‘The Oxford Consultation on Copyright and AI’) earlier this month, drawing on input from the creative, media and tech industries. Despite contributors’ differing views, the report is said to represent a “broad consensus arrived at over many months”.

Although the report focuses on the threats AI poses to the UK’s creative sector and the solutions needed to address them, its recommendations have wider relevance.

The recommendations are organised around three pillars (or principles), outlined below.

Ownership and content

The recommendations under the first pillar (or principle), ownership and content, are:

• Publicly recognise that using copyright-protected work for AI purposes requires the rights holder’s explicit authorisation, unless the rights holder has made the work available under open access or Creative Commons terms, or an existing copyright exception applies. Authorisation should be obtained through a legally binding licensing regime that enables rights holders to set permissions and preferences and choose whether to consent to use of their work. No new copyright exception or automatic opt-out should be introduced unless it upholds this approach.
• Review and, where needed, adapt and expand primary and secondary copyright infringement legislation to address how copyright-protected works are used in an AI context (citing the first-instance UK decision in Getty Images v Stability AI as an example). A related recommendation is that any AI-related product, service or model imported into or made available in the UK must have the relevant permissions to use UK copyright-protected work in the UK market, even if that work was copied or scraped outside the UK.
• Retain current copyright exceptions.
• Introduce a personality right to protect the use of a person’s name, image or likeness (covering use of a person’s face, voice, body, or artistic performances). Interestingly, the report proposes extending copyright to achieve this rather than introducing a separate right. It also suggests that “creators should also have appropriate protection against the unauthorised commercial exploitation of their distinctive style”.

Transparency, traceability and disclosure

The recommendations under the second pillar (or principle), transparency, traceability and disclosure, are:

• Enable traceability of training data and source material by setting minimum standards for data provenance and licensing information, ensuring licensing information can be embedded as standard, and prohibiting its removal or alteration.
• Require so-called ‘ingredient transparency’. AI developers would need to keep and, subject to certain exemptions (see below), publish detailed records of the data used to train AI models and any source material contributing to AI-generated content, excluding weights, personal prompts and data.
• Require minimum standards of interoperability (rather than trying to prescribe a single technical solution) to enable rights holders, regulators and other interested parties to scan and compare the metadata of different AI models and products.
• Establish a public register listing all web-crawling/web-scraping bots, their stated purpose, and the responsible entity. Require that all (web-crawling/web-scraping) bots hold cryptographic proof of identity (enabling web domains to distinguish between human and AI-related traffic).
• Require AI-generated works to be labelled, so that individuals know when they are interacting with or exposed to AI-generated content. (This reflects the existing transparency requirements in the EU’s AI Act.)
• Establish a simple and easily understandable way for rights holders and individuals with personality rights to see if and how their copyright-protected work or their name, image or likeness have been used for AI purposes.
• Allow limited exemptions from transparency requirements for: (a) national security, (b) specialised systems trained on sensitive datasets, and (c) (in exceptional cases) for commercial sensitivity, provided there is still full regulatory access on a confidential basis.

Fair licensing and market mechanisms

The recommendations under the third pillar (or principle), fair licensing and market mechanisms, are:

• Avoid introducing a centrally administered ‘levy system’ (discussed in the UK government’s report on copyright and AI earlier this year) or building and operating a licensing marketplace through the proposed Creative Content Exchange. Instead, establish and enforce the conditions for an independent, competitive licensing market by implementing the report’s recommendations and allowing a market for the lawful, fair and efficient licensing of copyright-protected work for AI purposes to develop organically.
• Support the development of three complementary licensing models (with the rights holder to determine the applicable model for its copyright-protected work): (A) data banks and combined datasets; (B) collective licensing; and (C) direct negotiation.
• Establish a UK government-led pilot, before introducing any mandatory requirements, to test whether structured, licensed, machine-readable access to publicly available datasets (excluding highly valuable, sensitive or commercially strategic datasets) can support AI developers’ transparency and compliance obligations.

In its report, the Institute for Ethics in AI calls on the UK government to:

• Move quickly to implement the recommendations in the report.
• Ensure that each central and local government department’s procurement, development and use of AI models comply with the report’s recommendations and principles.
• Develop the UK’s policy on AI and copyright in close consultation with rights holders (including creators, creative organisations and relevant unions).
• Introduce new legislation to implement or support the recommendations in this report, including establishing offences and meaningful penalties for non-compliance, and creating statutory penalties for AI-related copyright breaches (without requiring a rights holder to pursue copyright infringement litigation).
• Designate a regulator with primary responsibility for regulating and enforcing the recommended measures, including responsibility for enforcing AI-related copyright infringements and statutory penalties. This could be a new or existing regulator, but it would need the authority, independence and resources to require compliance and exercise its enforcement powers, as well as the confidence of the creative and technology sectors. The report suggests this would rule out the Intellectual Property Office and the Information Commission’s Office.
• Incentivise compliance through whitelisting mechanisms. These could make compliant firms eligible for government grants, innovation funding, procurement processes, tax incentives, regulatory sandboxes or access to public data sources. The report also suggests that firms complying with the three pillars could qualify for a ‘responsible AI use’ certification.
• Support two routes for copyright enforcement: traditional infringement litigation through the courts (with the report also proposing a fast-track dispute resolution process), and a new route allowing the regulator to take enforcement action against repeat offenders, with fines based on a penalty-per-breach principle.

Next steps

The Institute for Ethics in AI is clearly critical of the UK government’s action to date. The report acknowledges that the government has “repeatedly stated that copyright law must be upheld”, but argues that, in practice, it “has stood in the way of steps that would provide copyright holders with the means to uphold the law, through transparency and labelling”.

The report is the latest contribution to the wider debate on the UK’s AI and copyright policy. It urges the government to implement its recommendations quickly, highlighting potential rewards and opportunities for the UK, including becoming a global leader in AI-related copyright licensing. However, it remains to be seen whether, and to what extent, the government will act on them.