Widespread concern about the use of creative works to train artificial intelligence (AI) systems has prompted the UK government to begin exploring how the country’s copyright rules can be changed to satisfy the complex, often conflicting demands of both the creative and tech sectors.
The options being assessed include keeping copyright and related laws remain as they are; strengthening copyright to require licenses in all cases; implementing a broad data mining exemption for AI companies; or creating a more limited data mining exemption that allows copyright holders to reserve their rights, underpinned by measures to promote and support greater transparency from developers.
However, given structural imbalances within existing copyright markets – which favour giant corporations over individual creators – it is unclear to what extent the AI-related reforms to the UK intellectual property rules being considered will help creative workers themselves.
Creators vs AI developers
Questions around the use of creative works to train AI systems have become one of the most intense areas of debate since the advent of generative AI (GenAI) and large language models (LLMs) with the release of OpenAI’s ChatGPT in November 2022.
Aside from a lack of transparency from AI companies about the data included in their training corpuses, creatives have variously complained about the absence of enforceable mechanisms to protect their copyrighted works within the context of scraping at scale, as well as the impacts of AI on creative job markets and competition.
For AI companies, on the other hand, access to vast amounts of high-quality data is of paramount importance, particularly when it comes to the development of LLMs such as Claude, ChatGPT or Gemini.
A submission to the US Copyright Office on 30 October 2023 by Amazon and Google-backed LLM developer Anthropic is indicative of how these firms view their use of copyrighted material, and how integral it is for creating generative AI models.
“To the extent copyrighted works are used in training data, it is for analysis (of statistical relationships between words and concepts) that is unrelated to any expressive purpose of the work,” it said. “This sort of transformative use has been recognised as lawful in the past and should continue to be considered lawful in this case.”
It added that using copyrighted works to train its Claude model would count as “fair use”, because “it does not prevent the sale of the original works, and, even where commercial, is still sufficiently transformative”.
As part of a separate legal case brought against Anthropic by major music publishers in November 2023, the firm took the argument further, claiming “it would not be possible to amass sufficient content to train an LLM like Claude in arm’s-length licensing transactions, at any price”.
It added that Anthropic is not alone in using data “broadly assembled from the publicly available internet”, and that “in practice, there is no other way to amass a training corpus with the scale and diversity necessary to train a complex LLM with a broad understanding of human language and the world in general”.
If licences were required to train LLMs on copyrighted content, today’s general-purpose AI tools simply could not exist Anthropic
“Any inclusion of plaintiffs’ song lyrics – or other content reflected in those datasets – would simply be a byproduct of the only viable approach to solving that technical challenge,” it said.
It further claimed that the scale of the datasets required to train LLMs is simply too large to for an effective licensing regime to operate: “One could not enter licensing transactions with enough rights owners to cover the billions of texts necessary to yield the trillions of tokens that general-purpose LLMs require for proper training. If licences were required to train LLMs on copyrighted content, today’s general-purpose AI tools simply could not exist.”
While the submission and the court case are specific to the US context, the application of “fair use” exemptions to copyright is not dissimilar in UK. Under current UK copyright laws, original works are automatically protected upon their creation, giving the creators exclusive rights to copy, distribute, perform or adapt their work.
There are, however, limited exemptions that allow the “fair dealing” of copyrighted material for the purposes of, for example, research, criticism, review and reporting. A further exemption was added in 2014, allowing text and data mining for purely non-commercial research purposes.
As it stands, unless one of these exemptions applies, AI companies would therefore need to obtain permission from copyright holders to use these works in their model’s training data.
UK government consultation backlash
According to a previous UK government consultation on the matter, which closed in February 2025, “the application of UK copyright law to the training of AI models is disputed”.
It said that while rights holders are finding it difficult to control the use of their works in training AI models, and are seeking to be remunerated for its use, AI developers are similarly finding it difficult to navigate copyright law in the UK. It noted “this legal uncertainty is undermining investment in and adoption of AI technology”.
In an attempt to solve the dispute, the UK government proposed a new policy in late 2024 that would allow AI companies to train their models on copyrighted works unless rights holders explicitly opted out. This means that, rather than requiring AI companies to seek permission from rights holders for the use of their work, the burden would be placed on the creators themselves to actively object.
The opt-out proposal provoked significant backlash from creatives, who viewed it as too conciliatory to the narrow interests of tech companies. Out of the more than 10,000 people that responded to the government’s consultation on these measures, just 3% backed it’s opt-out proposal, while 95% called for either called for copyright to be strengthened, a requirement for licensing in all cases, or no change to current copyright law.
Others cited issues around the practicality of such proposals, noting that in the context of the current digital landscape – where copyrighted content is scraped at scale and included in training datasets, often without attribution – it may be impossible for someone to know when their work has been used, let alone opt out.
In the wake of this widespread opposition, the UK government has since committed to exploring a licence-first system that would require AI companies to seek explicit permission from creatives and provide them with compensation.
Balancing interests?
A year later, in December 2025, technology secretary Liz Kendall told Parliament there was “no clear consensus” on the AI-copyright issue, saying that the government would “take the time to get this right” while promising to make policy proposals by 18 March 2026.
“Our approach to copyright and AI must support prosperity for all UK citizens, and drive innovation and growth for sectors across the economy, including the creative industries,” she said. “This means keeping the UK at the cutting edge of science and technology so UK citizens can benefit from major breakthroughs, transformative innovation and greater prosperity. It also means continuing to support our creative industries, which make a huge economic contribution, shape our national identity and give us a unique position on the world stage.”
While government rhetoric on AI and copyright has revolved around the need to support both the UK’s creative and tech sectors, there is a sense that – so far at least – it is prioritising the latter in its ambition to make the country a tech superpower.
Beeban Kidron, a crossbench peer and former film director, for example, has previously described the use of copyrighted material by AI companies as “state-sanctioned theft”, claiming ministers would be “knowingly throwing UK designers, artists, authors, musicians, media and nascent AI companies under the bus” if they don’t move to protect their output from being harvested by AI firms.
Owen Meredith, chief executive of the New Media Association, has also previously urged the UK government to rule out any new copyright exception. “This will send a clear message to AI developers that they must enter into licensing agreements with the UK’s media and creative copyright owners, unlocking investment and strengthening the market for the high-quality content that is the most valuable ingredient in producing safe, trustworthy AI models,” he said.
Ed Newton-Rex, a prominent commentator on AI and intellectual property, has also criticised the balance of UK government’s approach, noting that while the government described its consultation proposals at the time as a “win-win … this is very far from the truth. It would be a huge coup for AI companies, and the most damaging legislation for the creative industries in decades”.
He added that a broad copyright exception that allows unlicensed training on copyrighted “would hand the life’s work of the UK’s creators to AI companies, letting them use it to build highly scalable competitors to those creators with impunity”.
[A broad copyright exception] would hand the life’s work of the UK’s creators to AI companies, letting them use it to build highly scalable competitors to those creators with impunity Ed Newton-Rex, AI and intellectual property commentator
AI companies, of course, disagree. In its October 2023 submission to the US Copyright Office, Anthropic argued that requiring licences would be inappropriate, as it would lock up access to the vast majority of works and benefit “only the most highly resourced entities” that are able to pay their way into compliance
“Requiring a licence for non-expressive use of copyrighted works to train LLMs effectively means impeding use of ideas, facts and other non-copyrightable material,” it said. “Even assuming that aspects of the dataset may provide greater ‘weight’ to a particular output than others, the model is more than the sum of its parts. Thus, it will be difficult to set a royalty rate that is meaningful to individual creators without making it uneconomical to develop generative AI models in the first place.”
Others from the tech sector have also argued that diverging from other jurisdictions too greatly – for example, by implementing a UK-specific licensing arrangement preferred by the creative sector, or requiring firms to disclose detailed data inputs – would simply mean AI companies avoid deploying in the UK.
Trade association TechUK, for example, argued that in the context of AI-copyright related amendments to the government Data Use and Access Bill – which would have forced developers to publish their training corpuses but which were ultimately not included in the final Act of Parliament – departing too much from existing UK and international frameworks would risks companies being “discouraged from operating, training and deploying AI products and models in the UK”.
This was also recognised by the government in its consultation, which noted requiring licenses in all cases “is highly likely to make the UK significantly less competitive compared to other jurisdictions – such as the EU and US – which do not have such restrictive laws. This would make the UK a less attractive location for AI development, reducing investment in the sector. In doing so, it may not actually increase the level of licensing undertaken by AI firms.”
It added that models trained in other jurisdictions which do not meet any new UK standards may be difficult to restrict from the UK market, and risks some of the most capable AI models not being made available in the UK: “This would significantly limit innovation, consumer choice and wider benefits of AI adoptions across the UK economy.”
The technical caveats of copyright law
Under UK copyright law, it should be noted that creating “transient copies” of works is allowed if certain conditions are met. This includes if it’s not a permanent copy and serves a brief, ancillary purpose; if it’s a necessary step in a technological process; if its only goal is enabling lawful use or network transmission; and the copy itself doesn’t hold separate commercial value.
When looking at AI model training processes – which often, but not always, retain only a very small portion of each training item – this indicates it would be technically wrong to assert a copyright infringement has taken place, as Anthropic has argued in the context of the US.
However, this doesn’t mean that a model would never infringe copyright, as it is also technically possible for most models to “memorise” copyrighted works, turning a transitory copy into a permanent, infringing one.
Although the specificities of whether a particular model or AI-generated output infringes this current copyright regime will be hashed out in individual court cases, some have argued that looking for copyright to solve the tension between creatives and AI companies is a non-starter.
In their book Chokepoint capitalism: How big tech and big content captured creative labor markets and how we’ll win them back, for example, authors Cory Doctorow and Rebecca Giblin argue that while the past 40 years have been spent elaborating international copyright rules, the financial benefits of this have largely accrued to big business rather than creators themselves, whose share of growing entertainment industry profits have declined in that time.
In essence, their argument is that expanding copyright is very unlikely to protect the jobs or incomes of already underpaid creatives, who have themselves been exploited by entertainment behemoths wielding copyright laws against them for decades.
In their May 2024 book, Who owns this sentence? A history of copyright and wrongs, authors Alexandre Montagu and David Bellos similarly argue that copyright protections – which were originally intended to protect the livelihoods of individual creators – have since been transferred to giant corporations instead, which use them to extract a form of “rent” from consumers globally, while also locking the employees who helped contribute to the creation of the IP out from ownership and the consequent benefits.
It follows, then, that there is little reason to believe these same companies will now treat their creative workers more fairly if they receive compensation as a copyright holder from AI companies.
To alter this dynamic, Doctorow and others argue it would require changing the very structure of creatives markets so that the benefits accrue to creatives, rather than large corporations that essentially run “tollbooths” to facilitate and control access to creative’s work, which in turn allows them to extract disproportionally high profits for themselves.
Writing for the Electronic Frontier Foundation (EFF) in February 2025, Tori Noble argued that “expanding copyright will not mitigate” the harm to creative workers, and that “what neither Big Tech nor Big Media will say is that stronger antitrust rules and enforcement would be a much better solution”.
She added that looking beyond copyright can future-proof protections, including stronger environmental protections, comprehensive privacy laws, worker protections and media literacy, adding: “[This will] create an ecosystem where we will have defences against any new technology that might cause harm in those areas, not just generative AI. Expanding copyright, on the other hand, threatens socially beneficial uses of AI – for example, to conduct scientific research and generate new creative expression – without meaningfully addressing the harms.”
Collective copyright and labour law
As it currently stands, UK government looks to be on course to introduce a new licensing regime for AI companies’ use of copyrighted materials. Observers have said this would need to include mechanisms that allow creators to identify when and how their works are used, as well as to object or seek compensation as they see fit.
However, given the clear tensions that already exist between individual and corporate copyright holders, even a licensing regime could still disproportionally benefit the latter. It could also disproportionally benefit large AI developers, as the pool of actors with the ability to pay for enough copyright licenses to effectively train a model is vanishingly small.
The use of AI in creative endeavours throws up further issues around labour and competition: even if creators received compensation for the use of their copyrighted material, AI’s entire development is underpinned by a neoliberal logic of austerity. This means that, in the current political-economic context, those with the decision-making power to deploy AI largely do so because it allows them to cut labour costs – the biggest overhead for any capitalist enterprise.
In the current political-economic context, those with the decision-making power to deploy AI largely do so because it allows them to cut labour costs – the biggest overhead for any capitalist enterprise
In November 2024, data from the Harvard Business Review showed the impact that generative AI models were already having on labour markets, which highlights how creatives will essentially end up competing with the very models that ingest their data. Specifically, it highlighted how the introduction of ChatGPT decreased writing and coding jobs by 30% and 20% respectively, while AI image generators similarly decreased image creation jobs by 17%.
Given the sheer scale at which models ingest data, it is not hard to see how creatives – even with a licensing regime in place – could be undermined by bosses who would rather pay for a relatively cheap corporate licence to an AI model, rather than the comparatively expensive labour of human beings.
While many argue that the advent of AI is inevitable, its impacts are certainly not. In November 2023, for example, the Autonomy think tank in the UK argued that while automating jobs with LLMs could lead to significant reductions in working time without a loss of pay or productivity, realising the benefits of AI-driven productivity gains in this way will require concerted political action.
The think tank added this was because it is clear that productivity gains are not always shared evenly between employers and employees, and depend on “geographic, demographics, economic cycle and other intrinsic job market factors” such as workers’ access to collective bargaining.
To deliver positive AI-led changes for workers and not just employers, Autonomy recommended setting up “automation hubs”, underpinned by trade union and industry agreements, to boost the adoption of LLMs in ways that are equitable.
In the context of the creative industries and copyright, a similar situation has already taken place with the 2023 Hollywood writers’ strike, whose collective sector-wide action ended with an agreement from studios that AI cannot be used to write or rewrite scripts, and which gave them the ability to prohibit the use of their writing in model training.
Instead of replying purely on copyright law – which historically has been wielded against individual creatives by entertainment and media companies – the answer may be found in attempting to build up collective copyright mechanisms and improving the underlying labour protections for creative workers to stop them being ripped off by companies, with or without the help of AI.
Advanced Machine Intelligence (AMI), a new Paris-based startup cofounded by Meta’s former chief AI scientist Yann LeCun, announced Monday it has raised more than $1 billion to develop AI world models.
LeCun argues that most human reasoning is grounded in the physical world, not language, and that AI world models are necessary to develop true human-level intelligence. “The idea that you’re going to extend the capabilities of LLMs [large language models] to the point that they’re going to have human-level intelligence is complete nonsense,” he said in an interview with WIRED.
The financing, which values the startup at $3.5 billion, was co-led by investors such as Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions. Other notable backers include Mark Cuban, former Google CEO Eric Schmidt, and French billionaire and telecommunications executive Xavier Niel.
AMI (pronounced like the French word for friend) aims to build “a new breed of AI systems that understand the world, have persistent memory, can reason and plan, and are controllable and safe,” the company says in a press release. The startup says it will be global from day one, with offices in Paris, Montreal, Singapore, and New York, where LeCun will continue working as a New York University professor in addition to leading the startup. AMI will be the first commercial endeavor for LeCun since his departure from Meta in November 2025.
LeCun’s startup represents a bet against many of the world’s biggest AI labs like OpenAI, Anthropic, and even his former workplace, Meta, which believe that scaling up LLMs will eventually deliver AI systems with human-level intelligence or even superintelligence. LLMs have powered viral products such as ChatGPT and Claude Code, but LeCun has been one of the AI industry’s most prominent researchers speaking out about the limitations of these AI models. LeCun is well known for being outspoken, but as a pioneer of modern AI that won a Turing award back in 2018, his skepticism carries weight.
LeCun says AMI aims to work with companies in manufacturing, biomedical, robotics, and other industries that have lots of data. For example, he says AMI could build a realistic world model of an aircraft engine and work with the manufacturer to help them optimize for efficiency, minimize emissions, or ensure reliability.
AMI was cofounded by LeCun and several leaders he worked with at Meta, including the company’s former director of research science, Michael Rabbat; former vice president of Europe, Laurent Solly; and former senior director of AI research, Pascale Fung. Other cofounders include Alexandre LeBrun, former CEO of the AI health care startup Nabla, who will serve as AMI’s CEO, and Saining Xie, a former Google DeepMind researcher who will be the startup’s chief science officer.
The Case for World Models
LeCun does not dismiss the overall utility of LLMs. Rather, in his view, these AI models are simply the tech industry’s latest promising trend, and their success has created a “kind of delusion” among the people who build them. “It’s true that [LLMs] are becoming really good at generating code, and it’s true that they are probably going to become even more useful in a wide area of applications where code generation can help,” says LeCun. “That’s a lot of applications, but it’s not going to lead to human-level intelligence at all.”
LeCun has been working on world models for years inside of Meta, where he founded the company’s Fundamental AI Research lab, FAIR. But he’s now convinced his research is best done outside the social media giant. He says it’s become clear to him that the strongest applications of world models will be selling them to other enterprises, which doesn’t fit neatly into Meta’s core consumer business.
As AI world models like Meta’s Joint-Embedding Predictive Architecture (JEPA) became more sophisticated, “there was a reorientation of Meta’s strategy where it had to basically catch up with the industry on LLMs and kind of do the same thing that other LLM companies are doing, which is not my interest,” says LeCun. “So sometime in November, I went to see Mark Zuckerberg and told him. He’s always been very supportive of [world model research], but I told him I can do this faster, cheaper, and better outside of Meta. I can share the cost of development with other companies … His answer was, OK, we can work together.”
Nvidia is planning to launch an open-source platform for AI agents, people familiar with the company’s plans tell WIRED.
The chipmaker has been pitching the product, referred to as NemoClaw, to enterprise software companies. The platform will allow these companies to dispatch AI agents to perform tasks for their own workforces. Companies will be able to access the platform regardless of whether their products run on Nvidia’s chips, sources say.
The move comes as Nvidia prepares for its annual developer conference in San Jose next week. Ahead of the conference, Nvidia has reached out to companies including Salesforce, Cisco, Google, Adobe, and CrowdStrike to forge partnerships for the agent platform. It’s unclear whether these conversations have resulted in official partnerships. Since the platform is open source, it’s likely that partners would get free, early access in exchange for contributing to the project, sources say. Nvidia plans to offer security and privacy tools as part of this new open-source agent platform.
Nvidia did not respond to a request for comment. Representatives from Cisco, Google, Adobe, and CrowdStrike also did not respond to requests for comment. Salesforce did not provide a statement prior to publication.
Nvidia’s interest in agents comes as people are embracing “claws,” or open-source AI tools that run locally on a user’s machine and perform sequential tasks. Claws are often described as self-learning, in that they’re supposed to automatically improve over time. Earlier this year, an AI agent known as OpenClaw—which was first called Clawdbot, then Moltbot—captivated Silicon Valley due to its ability to run autonomously on personal computers and complete work tasks for users. OpenAI ended up acquiring the project and hiring the creator behind it.
OpenAI and Anthropic have made significant improvements in model reliability in recent years, but their chatbots still require hand-holding. Purpose-built AI agents or claws, on the other hand, are designed to execute multiple steps without as much human supervision.
The usage of claws within enterprise environments is controversial. WIRED previously reported that some tech companies, including Meta, have asked employees to refrain from using OpenClaw on their work computers, due to the unpredictability of the agents and potential security risks. Last month a Meta employee who oversees safety and alignment for the company’s AI lab publicly shared a story about an AI agent going rogue on her machine and mass deleting her emails.
For Nvidia, NemoClaw appears to be part of an effort to court enterprise software companies by offering additional layers of security for AI agents. It’s also another step in the company’s embrace of open-source AI models, part of a broader strategy to maintain its dominance in AI infrastructure at a time when leading AI labs are building their own custom chips. Nvidia’s software strategy until now has been heavily reliant on its CUDA platform, a famously proprietary system that locks developers into building software for Nvidia’s GPUs and has created a crucial “moat” for the company.
Last month The Wall Street Journal reported that Nvidia also plans to reveal a new chip system for inference computing at its developer conference. The system will incorporate a chip designed by the startup Groq, which Nvidia entered into a multibillion-dollar licensing agreement with late last year.
Paresh Dave and Maxwell Zeff contributed to this report.
Anthropic executives allege that current customers and prospective ones have been demanding new terms and even backing out of negotiations since the US Department of Defense labeled the AI startup a supply-chain risk late last month, according to court papers that also revealed new financial details about the company.
Hundreds of millions of dollars in expected revenue this year from work tied to the Pentagon is already at risk for Anthropic, the company’s chief financial officer, Krishna Rao, wrote in a court filing on Monday. But if the government has its way and pressures a broad range of companies from doing business with the AI startup, regardless of any ties to the military, Anthropic could ultimately lose billions of dollars in sales, he stated. Its all-time sales, since commercializing its technology in 2023, exceed $5 billion, according to Rao.
Anthropic’s revenue exploded as its Claude models began outperforming rivals and showing advanced capabilities in areas such as generating software code. But the company spends heavily on computing infrastructure and remains deeply unprofitable. Rao specified that Anthropic has spent over $10 billion to train and deploy its models.
Anthropic chief commercial officer Paul Smith provided several examples of partners who have privately raised concerns to the AI startup in recent days. He said a financial services customer paused negotiations over a $15 million deal because of the supply-chain label, and two leading financial services companies have refused to close deals valued together at $80 million unless they gain the right to unilaterally cancel their contracts for any reason. A grocery store chain canceled a sales meeting, citing the supply-chain-risk designation, Smith added.
“All have taken steps that reflect deep distrust and a growing fear of associating with Anthropic,” Smith wrote.
The executives’ comments are part of statements from six Anthropic leaders in support of a preliminary order that would allow the San Francisco company to continue doing business with the Department of Defense until lawsuits about the supply-chain-risk issue are resolved.
Anthropic has sued the Trump administration in two courts. A lawsuit filed in San Francisco federal court on Monday alleges the government violated the company’s free speech rights. A separate case filed Monday in the federal appeals court in Washington, DC, accuses the Defense Department of unfairly discriminating and retaliating against Anthropic.
The company is seeking a hearing as soon as Friday in San Francisco for a temporary reprieve. The legal battle and sales fallout follows a weeks-long dispute between Anthropic and the Pentagon over the potential use of AI technologies for mass domestic surveillance and autonomous lethal weapons. Anthropic contends AI is not yet capable of safely undertaking the tasks, while the Pentagon wants the right to make that judgment on its own.
By law, the supply-chain designation prevents a narrow set of companies that do business with the Pentagon from incorporating Anthropic into their systems. But Defense secretary Pete Hegseth has cast a wider net. He posted on X late last month that “effective immediately, no contractor, supplier, or partner that does business with the United States military may conduct any commercial activity with Anthropic.”
Rao wrote that the Pentagon reinforced the message by reaching out to several startups about their use of Claude, which he said he learned had happened from speaking with an investor that Anthropic and the smaller companies all share. They “have grown worried and uncertain about their ability to use Claude,” Rao wrote.
The Pentagon declined to comment on the lawsuits and did not immediately respond to a request for comment about Rao’s allegation about the outreach.
Politics3 days ago
Sports3 days ago
Entertainment3 days ago
Tech1 week ago
Sports1 week ago