The Sovereign AI Fund is refusing to answer a basic question around its policy on copyright.

Something funny is going on at the UK's new Sovereign AI Fund.

In this post, I will make the case that they are intentionally leaving the door open to invest UK taxpayer money in companies that train AI models on copyrighted work without a licence - and they may even have already done so.

Not answering the question

When someone won't answer a question you ask them, it is natural to ask why.

Last week, I started asking various people involved with the Sovereign AI Fund whether it would invest in companies that train on copyrighted work without a licence. Others asked them the same question.

Most didn't reply - only the chair of the fund, James Wise, did. Several times. But he never answered the question that had been asked.

He gave answers - several of them. But they were answers to different questions. The fact that he wouldn't answer the question I had asked is deeply concerning.

‘Following the law’

When James did answer, he said they would only invest in companies that comply with the law.

In his most detailed response, he said the following:

We will of course only invest in companies that follow the law on this issue. It is really important - and we have been clear ( many times! ) that copyright rules should be respected, use of copyright works to train AI in the UK will require a licence unless an exception exists.

For someone unfamiliar with the issues around AI and copyright, this may seem like he has answered the question. But he hasn't.

He adds two qualifications to his answer, and these qualifications change the meaning entirely.

First, he adds the words “in the UK”. In other words, he is only speaking about AI models trained in the UK - not models owned by a British company but trained on servers abroad.

Second, he adds the words “unless an exception exists”. In other words, he is not ruling out that some form of copyright exception may currently exist in the UK that applies to AI training, which would negate the need for the company to license its training data.

What does this leave open? Well, here is one specific example: if a UK company trained its AI models on GPUs in the US, it would not be addressed by James' answer, since US law on AI training is currently disputed. He has not ruled out the Sovereign AI Fund investing in such a company.

Have they already made investments like this?

Such has been their unwillingness to answer this simple question that it has led me to wonder whether they may already have companies that train on copyrighted work without a licence in their portfolio.

Looking at their initial portfolio, which they announced last week, this seems eminently possible. Why?

• It includes companies that build the kind of models that are often trained on copyrighted work without a licence. Not all models are trained this way, but many sadly are.

• It includes companies that seem to be based in both the US and the UK. And, as we know, many AI companies in the US train on copyrighted work without a licence.

• It has companies that don't reveal the source of their training data. In my experience, if an AI company has avoided training on copyrighted work without a licence, it is usually open about this.

Now, I don't know what most of these companies train on. I can't know, because they haven't revealed it. But, given the above, it is clearly possible that some of them train on copyrighted work without a licence.

In an industry in which training on copyrighted work without a licence is sadly all too common, it is not a huge stretch to think that some among this group of companies may do so.

Getting defensive

Another thing that has led me to conclude that the Sovereign AI Fund is intentionally leaving these possibilities open has been the way James has interacted with me.

I have asked questions politely, even in the face of him accusing me of acting in bad faith - and he has become even more defensive, resorting to insults and blocking me.

I hope that anyone who knows me or who follows me on Twitter will agree that I am not a troll. I campaign largely on one issue: the exploitation of creatives by AI companies. I do so, I think, in good faith, often by asking simple questions.

My questions about the Sovereign AI Fund are a good example of this. Indeed, similar questions are being asked by members of the House of Lords this week:

This is not "trolling” - it is trying to get to the bottom of the Sovereign AI Fund’s policy. And, needless to say, I have not put out any “outrageous lie” - James does not point to one because one does not exist.

Refusing to answer the question they are being asked is already a bad sign. Things look even worse when they insult and block the people asking the question.

Why does this matter?

The root of the matter here is a simple question: what kinds of companies should UK taxpayer money be invested in?

Much of AI policy work is complicated by the global nature of the AI industry. As I wrote in a short policy paper last week:

The government has pointed out that it has no say over other countries' copyright laws, and no power to stop UK rights holders' work being used by AI companies for training abroad. It has no power, for instance, to stop OpenAI training on UK authors' work in data centres located in the US.

However, this issue vanishes with the Sovereign AI Fund. The Sovereign AI Fund invests in UK companies, rather than overseas companies. Decisions made by the Sovereign AI Fund have no bearing on overseas companies, and therefore raise no issues of extra-territoriality.

With the Sovereign AI Fund, our hands are not tied by the laws of other countries. We can choose the kinds of companies to support, and the kind of AI industry we want to promote. What kind of AI industry is that? Government ministers themselves have put it well. Also from my paper:

The government wants to support the emerging market for licensing AI training data. “We want more licensing of copyrighted material, not less,” said Peter Kyle in January 2025. “People rightly want to get paid for the work that they do,” said Liz Kendall, his replacement as Secretary of State for Science, Innovation and Technology, in November that year.

We can choose not to invest taxpayer money in companies that get around UK copyright law by simply training their models abroad. As the House of Lords Communications and Digital Committee recently put it, the UK can become “a world-leading home for responsible, licensing-based artificial intelligence (AI) development”. But only if we make the right policy choices.

So what can be done?

The paper I wrote last week, written before the Sovereign AI Fund launched, outlined a simple proposal that could align the incentives of the AI industry and rights holders here: the fund could commit not to invest in companies that train and copyrighted work without a licence, and could introduce a simple due diligence step to ensure this.

I still believe this is the best way forward. It is hard to imagine what the argument against it could be.

James seems frustrated by my persistence in asking what the sovereign AI fund's policy is here.

So my offer to James is that I will stop asking the question if he will answer it with a simple yes or no:

Will the Sovereign AI Fund invest in companies that train on copyrighted work without a licence?

There is no one who would be happier than me to stop having to ask this.

Edit: James later deleted his tweet calling me a troll and unblocked me. I have no idea what led to his change of heart.