Welcome to this final panel in our inquiry on artificial intelligence. Thank you very much to you all, Ministers, for coming along on this hot afternoon to give us evidence. We are really grateful to you. Kanishka, if I could start with you, we have seen some very different philosophical models of growth for artificial intelligence. We have seen a particular approach from the US. We have studied this in India. We have some experience of the way that China is approaching this. How would you characterise the UK’s theory of growth when it comes to AI?

Thank you, Chair. Can I first seek your permission to mirror you and take off my jacket?

Of course, yes. I am afraid Mr Speaker will insist that you keep your tie on. Jackets can be dispensed with.

I will very happily do so. I appreciate the generous start. At a very high level, there are two or three central choices that countries can make when they think about a vision of growth and its relation to artificial intelligence. One is the question of the extent to which you want to focus on being at the frontier of invention rather than the long tail of adoption. The second choice, which is really critical, is the extent to which adoption is primarily focused on particular parts of industry, with less of a focus on spread. The choice is between a focus on the absolute scale of adoption and on the spread and the qualitative nature of it. If I take those two dimensions, the broad read that I have is that the United States, for example, is extremely focused on the first question of the frontier and a bit less on adoption. Countries such as India, from my anecdotal experience, have been very focused on the development of small, application-focused models that can be put in the hands of millions and billions of people. Its focus is much more on the long tail of diffusion rather than the frontier of invention. Where does the UK sit on that question? Effectively, we are taking a very targeted approach. We are saying, “There are some very selective areas of frontier invention that are really critical to both productivity and national security, and we do not want to give up on those.” We are focused in those areas and we can talk about what those are in a second. At the same time, the bulk of our focus has been on driving adoption and diffusion. The second question is, “How do you think about the scale and scope of diffusion?” Effectively, in Britain, we have taken an approach where, of course, we want to move the needle in aggregate on productivity, but we want to do so by taking the country with us together. My belief and the Government’s belief is that you can only build a durable consensus behind technology in this wave, unlike the last, if the upside of that is spread across places and people as well. That really underpins a huge amount of the approach that we have taken to both infrastructure and wider AI adoption.

It sounds, therefore, like the model is a bit of both.

Yes, but it is really important for me to point out that it is not a hedge, because it is very targeted in both approaches. To take the frontier, if you said you were doing both the frontier and adoption, people might say that you were doing neither, if you had no priorities. On the frontier, our priorities are very surgical and targeted. For example, we are not trying to spend a huge amount of public money developing a frontier large language model. We think there are lots of non-language forms of intelligence, such as computer vision, world models, mathematical discovery, drug discovery or materials discovery. Britain has remarkable, unparalleled strengths in those, so we are very targeted on those areas of the frontier when it comes to invention. Similarly, on adoption, we are taking a very cluster and industry-focused approach, very much mirroring the IS-8 sectors in the industrial strategy.

One consequence of that is that, where we are not at the inventive frontier, we may be relying on others. The FT reported on 2 July that the new Prime Minister’s team was acutely concerned about a US-centric model characterised by “unfettered tech boosterism”. I do not want you to second-guess the new Prime Minister’s team, but I would be interested in your views about how the model of AI for this economy ought perhaps to change over the course of this Parliament.

Chair, perhaps you will forgive me for focusing more on a diagnosis of the present, rather than on a forecast of the future. On the present, the two big qualitative aspects that I and the Secretary of State have tried to bring to this role are, first, a shift in gear in focusing on small tech rather than big tech and, second, a laser-sharp focus on British sovereignty. I genuinely believe that what we have in Britain as a result is an approach to AI sovereignty that is distinctive. I am very happy to go into the detail of that, but the broad thrust of it is that Britain, given our scale and scope strengths, is best placed to build leverage in the stack rather than to build the full part of the AI stack entirely domestically. You have very significant compromises in the level of security that you have, if you are compromising on being at the frontier or buying the frontier for, say, defence purposes, or even aspects of the way that public service professionals are supported.

If we are not seeking to master the entire frontier of invention, there will be aspects of the frontier where we are going to rely on the United States. What are the options for strengthening our tech sovereignty and de-risking our relationship with the United States, which, as we know, has delayed our access to particular advanced models in recent weeks?

I am going to take a slightly longer approach to answering this. Let me first back out and say, “What is the point of sovereignty? What are the outcomes that determine whether you are sovereign or not?” To me, there are two core outcomes that I am really interested in that have defined our approach to British sovereignty on AI. The first outcome is that you need to be sovereign in having access to critical inputs that are core to your operation as a country. If you cannot have Nvidia chips, you cannot do inference workloads in AI and, as a result, you are not truly sovereign. We are not going, overnight or even over a multi-year period, to build Nvidia chips in this country. One definition of outcome on sovereignty is, “Can you access core inputs?” A second definition, complementary to it, is, “Can you shape this in the light of your values?” I start with those outcomes because that defines how you get there as a country with the scale and strength that Britain has. To me, there are three parts that determine whether you can achieve those outcomes. I call this broadly a ladder of strategic leverage that we have to build. The first part is that, for core inputs that you cannot develop in this country, you have to get enough of them today. I mentioned Nvidia chips. One of the first things that I did in this job was to go and negotiate for 100,000 GPUs, the largest deployment of GPUs in Europe, to be in the UK. Whatever you want to build—if you want to have national security deployments that are built in Britain—you can only do that with these chips. Securing enough is the first step of the ladder. Without that, you cannot go anywhere else. The second step is not to secure it from one supplier but to try to diversify. The core insight of Labour party politics—that bargaining power and who has power matters as much as what you do with it—is deployed in our strategy here. I went to see Groq, Cerebras and SambaNova and got all those other chip companies, including British chip companies, to accelerate their engagement with Britain so that over time we are diversifying where we are securing supply from and building more leverage for this country. The third step is, of course, in targeted areas, to build full-fat British. We have a very specific set of areas, such as inference chip design, which is where the future of chips is going; the series of non-language frontier intelligence models that I described earlier; and some very core areas of application, including life sciences, cyber-security, the services areas and national security, alongside trust and safety. That is effectively the package that we are building altogether in achieving the two outcomes that I described.

To sum up, it sounds like you are saying, “De-risk the supply critical inputs, diversify and become indispensable.” Those are the options for strengthening tech sovereignty over the course of this Parliament. What other consequences might there be for how we de-risk our relationship with the United States?

What are the consequences? If the question is about the impact of us successfully playing out the strategy that I have described, which effectively means that Britain is able to have much more leverage in the next five years than it had in the last 15 to 20 years, the consequences of that would be that we would have more ability to shape technology in the light of our values. It would mean that we create prosperity. As you will know better than I do, a major driver of Britain’s productivity position is the relative lack of capital depth in ICT compared to the United States. If we get this right, we will be richer as a country. If we are richer as a country, we will not only be richer in economic terms, but we will be richer in values terms because we will have shaped it here and built leverage for ourselves in where we are deploying it and how we are looking at the impact of it. Ultimately, we will be shaping it in the light of values that are core to our country.

Are there any implications for the way that we therefore have to rethink AI growth zones? Many of the beneficiaries of those AI growth zones have been American companies. Colleagues such as Baroness Kidron have argued that it is a huge benefit for American companies, which are enjoying access to cut-price electricity, for example. Does the strategy for AI growth zones have to change or evolve, or should we be pursuing the policy as it is?

As it relates to the conversation that we have just had, AI growth zones are a really critical part of Britain building its sovereignty. I say that in light of the ladder that I have described. We do not have the requisite level of compute in this country. We are not trying to blow the lights out. As you will know, the United States has 75% of data centre capacity globally. Britain barely features on the table. We have not built very much compute at all. If you do not build the compute in this country, you cannot build home-grown companies that are in the NHS. You cannot build home-grown companies at the heart of national security. You cannot build home-grown companies driving industrial and operational technology productivity, which I know you have heard from companies on as well. To me, this is the foundational layer, which is the first step on the ladder. In some cases, you have to get inputs from somewhere. The reason that most of the value in compute and data centres goes to the United States, and ultimately in large part to Taiwan and Korea, is that the chip companies that are the heart of it economically—70% to 75% of the value of data centres accrues to chips—are elsewhere. In the short term, we have to do that because that is the foundation on which we build sovereignty in other parts of the stack. Alongside that, as you will know, we launched this hardware plan. In three to five years’ time, I do not want us to be still sitting here and building data centres with foreign chips. The PUC is going towards more and more inference rather than training. Britain can make those chips. In fact, I think our best bet for reindustrialising is, using the power of private investment, to build chips in south Wales, in Edinburgh and Scotland, and parts of the north and Yorkshire, all of which have strengths in chips.

So you defend the current policy on AI growth zones as pretty integral to that wider vision that you have.

It is foundational. Without the compute layer and the plan to allow us to build chips and reindustrialise this country, I do not see how Britain builds sovereignty.

Much of the finance for this has been provided by Government centrally. A couple of days ago, the FT reported officials in your Department saying, off the record, that everyone is “mostly terrified” of Burnham’s plan to devolve powers. Is there a conflict between a broad shift towards devolving economic powers and resources, and the pursuit of the plan that you have just sketched out?

One of the reasons I try not to comment on anonymous briefings is that I find them mostly inaccurate. I find them inaccurate on two counts here. First, it is not the case, if the AI growth zone programme is the focus of this particular question, that most of that has been driven by public financing; in fact, almost none of it has been public financing. Through proper supply-side support, such as permitting and planning acceleration and rationalisation, the public sector has supported a very significant amount—in this case, £28 billion-plus—of private investment. There has been very little public investment that has driven that programme. Secondly, this does not run counter to the spirit of devolution; in fact, it only affirms it. The nature of these programmes was that these were plans and bids developed by local mayors alongside local authorities, the local university, local public services and, yes, in some cases, local private companies—for example, DataVita, a brilliant Scottish provider—and the international capabilities that other firms brought. We have a very cluster-specific approach in the north-east, driven by the strength of financial services in that area, anchored by SAGE and working with the mayors and local authorities. In Scotland, it was driven very much by the semiconductor industry, anchoring the rationality for why data centres are needed. In south Wales, again, there is the compound semiconductor cluster, and the work is driven by the Cardiff capital region primarily. In north Wales, the aspiration of clean energy and small modular reactors is at the heart of why data centres are being built there. Culham is the home of our nuclear fusion research agency, a public sector organisation. At the heart of this is a very cluster-specific and devolved approach. That has been the driving force behind the AI greats.

That is very helpful. Thank you for clearing that up.

This is a central question. I read that the Pope said that technology ought to remain human. In the realm of technology, my personal motto is to focus consistently on machines that serve us, rather than the other way around. If that is your primary lens to think about this question, there are lots of ways in which we have to be very clear in our choices to support that vision. One choice is who makes this technology. That is one of the reasons that I have been really passionate, with the Secretary of State, about making sure that, in this wave, unlike the last, the opportunity of fundamentally making this technology is in the hands of millions and millions of people right across this country—women, people from ethnic minority backgrounds, and kids growing up in places where prior generations might not have had that opportunity. What are we doing? We are spreading opportunity, as you will have seen, not least through the major charter that we have now signed with a series of the largest academic organisations. Women will have appropriate rights in the workplace when they are doing research jobs. That is a landmark advance in opportunity for women. What we have done on open source is another part of it. I am really proud that Britain led the last wave of open-source software, which means that a kid growing up in, say, Barry in my constituency is able to develop software on a Raspberry Pi at almost no cost and can then play a major role in shaping the future of technology. I want that to happen in AI as well.

In the last wave of technology, software was made in one of two ways. It could be closed source, which effectively meant that primarily large companies developed the software and retained all the IP. It was a shut ecosystem. To be able to participate in it and build new applications with that software, you had to purchase the IP and the licence. There was no way that a child sitting in south Wales without the economic potential could have done so. There was no way, by the way, that someone sitting in rural India or Africa could have done so. Alongside that, there was an open source movement. The web, for example, was open source. These were protocols that said, “We are not focused on shutting the IP. We are focused on building an ecosystem.” This was about letting one thousand flowers bloom and allowing people to create on top of that without having to pay for the licence. It was phenomenal—Britain was on the leading frontier of that—in terms of the sheer scale of open source talent. A child could now, for a few pounds, build software on a small Raspberry Pi, learn what that looks like and build applications that were remarkably useful. That meant a huge spread and diffusion of technology in the last wave. If we can capture some of that opportunity in this wave, we will have done a service not just to adoption but to the openness of adoption to people who otherwise would not be part of it.

Just following on from that, that is the very high-level, broad-sweep kind of thing. Blair, could I come to you to try to drill down a little bit into what it would mean for the economy? Where do the Government see the opportunities for growth? What sectors are going to benefit from this? What is the overall prize of driving AI across industry and business as a whole? Britain has a productivity problem; it is not uncommon in the west. Is AI the answer to this?

It can be an answer. It is one of the answers, certainly, to the productivity challenge that we have. I have the industrial strategy brief within my portfolio, but small and medium-sized enterprises are another particular brief. Kanishka has spoken about the wider industrial strategy element of this and making sure that the benefits of AI are spread around the country, where AI is an industry in itself. On top of that, we also have a job to do on the adoption of AI across the economy as a tool for addressing the productivity problem that we have. The third element—this is where I particularly focus on SMEs—is that we need to make sure that there is an equitable adoption of AI. Kanishka was describing previous waves of technology. One of the things that happened in the first wave of online shopping, for example, is that the opportunities were gobbled up by very large companies. Smaller traders were often last to the bowl on that. We have to make sure that the opportunities from agentic trading, for example, are spread equitably across the economy.

One of the things that we have heard today is that Government should be bold. When it comes to policy in this area, we need to be bold and not tinker around the edges. There is a balance to be struck here, isn’t there? It is very clear that AI is a good servant and a bad master. How can we be bold without being reckless?

It is not just a case of being bold without being reckless. Mr Aldridge was raising questions about the ethical issues and the regulatory side of things that DSIT oversees. We have to recognise that there is increasingly widespread adoption of AI at the moment. We can be very bold with our policies around encouraging the adoption of AI, but the evidence at the moment suggests that it is often quite widespread but quite shallow. People are experimenting with it. They are trying to upskill and learn how to use it as effectively as possible. Frankly, there is also a trust issue. People are not sure exactly how far they want to go with it.

We went on tour, and we were in Newcastle University. I was very struck by a thing that the people there said, which was that they were approached by a household-name company that said to them, “Give us AI.” The university said, “What do you mean?” They had no clue what they meant; they did not know what they wanted. They had heard about AI; they just needed AI—“Give us AI.” How can Government help businesses identify the question they are asking and get AI to give them the answer? It seems to be the other way around at the moment—the digital cart is in front of the horse at the moment.

I completely agree. It is an incredibly confusing landscape at the moment. It is confusing at any one moment, but it is also not a static landscape. If you can get to grips with it this week, there will be a new model in a couple of months’ time that will blow what you understood out the water very quickly. Skills England may have mentioned it earlier today, but we had the latest of our tech adoption roundtables in No. 10, where we try to bring together the big companies that offer AI services to business, with the aim of trying to curate better what is available out there so that people can understand the opportunities available to them and understand not just the funding, support and skills landscape from Government, but the huge offer that is out there from tech companies to support them on this journey. We will try to stay on top of this, but we need to enlist the tech companies, which are living and breathing this every day, to help business on that journey.

I will come to you very quickly, Kate. If we pull off this trick of maximising productivity through AI, how can we ensure that it feeds through into rising wages and helping with the cost of living crisis?

That is the opportunity in front of us. The world of work continues to change as technology and people’s experience of it in the workplace continue to change. If used well, for businesses and their employees alike, it absolutely leads to better productivity, growth in wages, growth across our economy and flexibility at work. There is a really positive opportunity for us to grasp, if done well and deployed well. That is what is at the forefront of my mind when we are approaching this topic. Making sure this enhances workers and grows wages is a real, key priority for us as a Government. As we grow our economy and improve living standards all across the country, work is absolutely essential to that, particularly as tech changes and impacts people’s experience of work. That is the future that we see. Kanishka and Blair have touched on that trust element and the adoption piece in terms of businesses. Trust is really vital to the uptake of tech and how it is used in the workplace. If employers bring the workforce with them, there is a real win-win here. We can use it responsibly, grow our economy and use it productively, and businesses can really reap the benefits of it. For example, using AI tools can help a business provide a better service and increase productivity. The question of who is benefiting from it is right at the forefront. This is a key question for us in DBT and in my work in the wider employment rights space. Successful adoption has to bring workers with us. They have to benefit too.

Let me just check something. Kanishka talked about a diffusion aspiration. One of the IS-8 sectors is digital and technology. You then have the other seven sectors and you have swathes of the economy bedevilled by this long tail of very poor productivity growth. Are the Government particularly trying to encourage AI in order to create an industrial strategy sector that is booming and therefore generating jobs and growth? Is it about making sure that the IS-8 stays at the cutting edge? Or is it about fixing this productivity question? You might tell me that it is all three of those things, but that is a hell of an ask and we have to make some choices.

It is. You will start to see more of an answer to that question as we publish the jobs plans that we are unpacking out of the industrial strategy. Those plans are looking at workforce needs generally within the economy, but they are also looking at the impact of AI within those sectors. From the point of view of where Kate sits, from the worker’s perspective, you will have an answer to some of those questions there. You also saw the beginnings of answers to some of those questions from the reports that were published by the AI champions within the industrial strategy sectors. Coming out of that, my question to your question is that you see them focusing on very particular aspects within each of their sectors. It is not that you are trying to boil the ocean. They are looking, within those industrial strategy sectors, which we have identified as the sectors we think have the most growth potential, for the particular issues they need to focus on to unlock innovation.

What I am hearing is, “We are figuring it out,” which is a reasonable answer.

There is an element of that because this is such a rapidly growing moment. The things that the AI champions have spoken about are deeply connected to the programmes that we already have in place as well. It is not that we are starting from scratch, but it will evolve.

Kate, the productivity gap with the United States has grown enormously since 2010. Productivity growth and wage growth began to get disconnected in about 2004 and it has not quite recovered. I have not heard the story that underlies Mr Cooper’s question yet, which is about how we should be self-confident about AI diffusion as a way of potentially augmenting skills and raising productivity, so that that productivity gain can be shared with workers in the form of bigger pay packets. That is what you were tilting into. I have not heard a member of the Cabinet set that story out in a bold way that would assuage what we know is the predominant feeling among workers, which is that people are pretty frightened about what this is going to do.

We have a real opportunity to set the standard in the UK of how we—

Do you believe it, though? Do you believe that there is a story there?

We are absolutely creating one. I will talk a bit about some of the stuff that we are doing in my brief, particularly on worker voice and experience, to feed into that story of how Government respond and ensure that we are not only keeping pace with changing technology, but looking at rights and making sure that they are keeping pace with changing technology in the workplace as well. Worker voice is a key element of how technologies are used and adopted responsibly by employers. Technology is absolutely a key part of creating more rewarding jobs and good-quality work, which is all part of our wider Make Work Pay agenda and feeds absolutely into this question. For me, technology is also a part of supporting people going through jobs transitions and supporting young people into work. It is a part of how we look at the labour market in the 21st century, how work continues, and will always continue, to change, and how we keep pace in terms of the rights that are in my brief. On how we are looking at and exploring this, Kanishka will touch on lots of work in DSIT around the AI Economics Institute and that cross-Government working that is taking place. DBT is feeding into that in terms of the AI and future of work unit to make sure that, across this whole space, we are prioritising secure and good-quality work. In terms of the worker voice piece, absolutely central to all of this, in our digital transition and adoption, is working with employers and with trade unions and workers, particularly on the workplace monitoring side of things, which is one experience of AI and how it is used in the workplace. We are continuing that dialogue with businesses and workers. We have launched a conversation and a consultation about that to build that evidence base of how it is being used, so that we can ensure we have that running thread I have talked about: workers are absolutely benefiting from these gains, we have a smooth transition, we can accelerate adoption and we use best practice throughout. This consultation that we have launched will be really key to building that evidence base and understanding what is happening in workplaces today.

That consultation is coming.

Yes. It has launched.

Blair, you are in this fantastic position. You have a new Prime Minister coming in. As a talented, ambitious new Minister, you will be writing advice to the incoming regime about things that have gone well and things that need to be improved in the future. What are your reflections about how we need to improve the incentives to diffuse AI through the economy in a way that helps fix this productivity problem that has bedevilled us for so long?

Sometimes the conversation on this focuses on cost, and sometimes it focuses on support. Having those two conversations separately slightly misses the point. The experience from Made Smarter, when we had the particular tech element to it, was that people were not taking advantage of the cost incentives, because the support was not there to do it at the same time. What we were discussing at the tech adoption roundtable earlier was making sure that the pathways to getting the right sort of support that is relevant to your business needs are there. When you get that support, it has to be connected to applications that not only are affordable but do not raise the fear that you are tied into something that you cannot get out of. There is a question of interoperability and people’s ability to move around. People feel this is a big leap. If they feel it is an irreversible leap, they will not take it. The companies need to work together on interoperability.

Those pathways are still to come.

This was the third of the tech adoption roundtables that we have had. As I say, we are working to try to curate that offer by the end of the summer.

Kanishka, we are all in agreement that these AI models are extraordinarily powerful, developing extraordinarily quickly and offer an enormous force for good in many respects. I want to explore the downsides. There is the AI Security Institute, which is responsible for our safety. At the moment, it has no statutory authority to force large tech companies to give their models to the institute before they are released into the public domain. Therefore, the institute is not testing any of these models. We are entirely reliant on the good graces of the Americans to get it right. Is that a satisfactory state of affairs? If not, what do you propose we do about it?

Let me just describe the current situation. The question of whether legislative specification should be a mechanism of approving the situation is, to me, secondary to the outcome that we want. Legislation might be a part of it, but to me the central outcome that we want is for Britain to be pretty much the best place in the world for assuring the quality of models and their risks before deployment, such that we can mitigate the downsides. That is the outcome that we want to obtain. Two facts are incredibly salient to how we achieve that outcome. Fact one is that, to my knowledge, no country in the world, certainly outside the United States, currently has access to models pre-deployment as a result of statutory specification. Fact two is that the UK is the only country in the world, outside of the United States, that yet has access to closed, proprietary frontier models from Google DeepMind, Anthropic and OpenAI to be able to evaluate them pre-deployment and understand the risks.

Is that all the models or some of the models?

I do not know about the first generation of models, but for the recent models we have evaluated all the core frontier models across the companies that I mentioned. You have to ask yourself a question. The system is putting the UK in a distinct position. Regardless of the fact that none of these companies are in this country from an incorporation or training point of view, they are doing this because we have used state capacity as the driver of the outcome rather than regulatory specification. It is delivering the outcome that we are interested in. It is delivering it uniquely of any country in the world. What is the rationale for switching from that mechanism to an alternative mechanism? I am not saying that I have the answer on this, but the big thing that I would say is that we have been able to secure that very distinctive outcome, to protect the British public and secure our British interests, because of the capacity and capability that we have built by working with the companies. I am not at all ruling out that putting that on a footing of statute at some point might be a helpful improvement, but the reality is that the outcome is what matters. We are obtaining that outcome. Anything that changes our mechanism of obtaining that outcome ought to be evaluated in light of that.

For OpenAI’s GPT-5.5, the UK AISI was unable to verify the effectiveness of the final configuration. For Claude Fable 5 and Mythos 5, AISI’s own testing is producing only interim results from a compressed testing window. We are relying on the good graces of America. I accept that the rest of the world is too, but we are entirely dependent on that. The AISI is not sufficient, as per its own testing results, to do a full test on these things. Ultimately, we are completely reliant on the Americans to get it right. I am not suggesting that this is your or anybody else’s fault, but that is just the reality of where we sit. Six weeks ago—not now, but six weeks ago—Trump in the White House was saying, “Take the guardrails off. Let’s run with this.” That is a pretty insecure world to live in, because we do not know what he will do tomorrow.

First, I would dispute that AISI has not been able to evaluate the models. They are present in the safety cards. I am very happy to share with the Committee further detail on the pretty full set of evaluations that we have conducted on the models. Secondly, it is true that, as a result of the failures of the British state and, frankly, the British technology sector over the last 15 years, we did not have any of these companies here. Yes, that does limit the policy space that we have in order to achieve the objectives that we want to achieve. That is exactly why the obsession that I have in this job is to build British sovereignty. When you put those things together, we have maintained this very tenuous ability that we have, to access models to evaluate, as a result of capability, not statute. A switch away from the mechanism that has worked has to be evaluated on those grounds.

Fair enough. Just switching over, “AI Scenarios 2030” was put together by AISI, DSIT and other experts across Government, academia and industry. The report concludes, “AI could cause serious, potentially even existential harms, without government intervention.” That is from your Department, I understand. I will just read that again: “potentially even existential harms”. Those are not my words; they are from a Government report. What do you see as proportionate action to be taken, given that that warning has been given?

The scenario described is one of the scenarios, so I want to caveat this by saying that there is a range of different scenarios. The scenario of existential risk that you are describing in particular is a pretty low-probability scenario. With that caveat, I take it very seriously because the impact is so high that, even if the probability is low, you want to take it seriously. There are a couple of appropriate actions that we should take. First, you have to build extremely deep understanding of the models and the ways in which that risk might obtain. Yes, the AI Security Institute is looking at cyber‑risks and biological and chemical weapon construction risk but, at the heart of it, it is also looking at loss of control and alignment risks, which are the ones that relate to the existential harm that you are describing. AISI is, as you know very well, the only lab in the world that has been able to evaluate models on each of those risks. In doing so, it both supports the companies to safeguard models from those risks and, internally, shares that understanding with national security agencies and the National Cyber Security Centre to mitigate those risks for critical national infrastructure. That is one part of it. Secondly, the reality of where we are, as you may have heard me say, is that this is the central question for our domestic and national security. It is a fact that large-scale firms are now primarily exposed to cyber-risk from AI. It is a fact that warfare, whether in Ukraine or Iran, is primarily determined by the quality of your drone and counter-drone operating systems, which are increasingly underpinned by AI. Those facts mean that, on your question of what a proportionate response would be, the most proportionate response would be for Britain to measure up to the scale of the diagnosis with a scale of investment in AI and to make it central to our public conversation. Thirdly, the nature of how models are being developed, not just closed, licensed models in the United States but, increasingly, openly proliferated models in China, over which we have no statutory ambit, means that we cannot simply rely on assessing models; we have to rely on societal hardening and resilience much more broadly. That requires not just a whole‑of‑Government but a whole‑of‑society approach. There is a series of things that we have tried to do through the National Security Council to prompt that as well.

Last October, MI5 noted the risks of these systems. What comfort can you give us and the country that this Government are on top of it? This is existential, in your own Department’s words. The fact that we have second looks at these models does not give a lot of people much comfort.

I might provide some comfort, but I also do not want to provide complacency. The comfort that I would provide is that, with respect, we have a first look at the models well before deployment. This country is the only country in the world that has a lab that is publicly located—

We cannot stop those models going out, can we?

We cannot stop those models, but we also cannot stop models that are extremely capable coming out of China, which are now increasingly proliferated. The idea that stopping these models is a tractable pathway is pretty challenging. This country is the only country in the world that is able to build a deep understanding—the deepest understanding—of these models and their risks in the public sector with democratic accountability, to be able to protect our core infrastructure against those risks. That is the comfort that we can provide. The second thing to say, though, is that comfort is no grounds for complacency. As I mentioned, these risks are extremely severe. I consider them to be the single most important set of risks for our country. At the same time, these are risks that we increasingly cannot simply control and box up. We have to prepare for resilience and hardness against these risks, rather than total containment. We now have to be in a very different conversation publicly about understanding, mitigating and adapting, rather than simply containing.

This is for you, Mr McDougall. We had your boss, the Secretary of State, in yesterday, giving us some quite vivid examples of his frustrations with getting investment through the door and some of the barriers particularly in the areas of AI and automation. As a Committee, we are trying to understand whether there need to be changes to the regulatory regime to increase investment. Is this an issue that you are coming up against?

Consistently across the economy, we have challenges in getting investment in. The place where it is most pronounced in the UK is the scale-up gap and the ability to get financing in there. In the space of AI—we were just discussing this before we came in—there are some pretty encouraging figures about the amount of capital coming into the UK compared to everywhere else in Europe combined.

You are taking us into the investment world, but we really wanted to know more about regulation. One thing that the Secretary of State talked about was someone further down the chain not having the expertise and knowledge to deal with the new world of AI and automation. Is that something that needs a new regulatory approach?

It does. You will be aware of the Regulators (Growth Objective) Bill that we are bringing forward. We fully expect that the regulators’ uses of the sandboxing element of that legislation will feature AI pretty heavily. For example, I was talking earlier to some people involved in cancer research about how the sandboxing would potentially better enable the use of AI for clinical trials and to accelerate research in that space. You can think elsewhere about more traditionally industrial uses, with autonomous marine and us not having a regulatory framework for that. Sandboxing can enable us to provide regulatory certainty for that, which then leads to the type of investment that you are talking about.

Thinking about the five 2030 scenarios that Charlie has just mentioned, does the Department favour one of those? Is there going to be a different regulatory approach depending on which scenario we are aiming for?

My approach with the regulation is about trying to get that investment in, particularly on the scale-up side of things. I will let the DSIT Minister talk about the regulation of AI, because that is his part of it. Our focus is about unlocking regulation that encourages innovation and, in turn, allows investment into industry.

Can I just come to you, Ms Dearden? You gave a very positive view of the benefits to workers from AI, if it is done in a good way. We had some evidence from the first panel about some of the challenges, particularly with algorithmic decision making in terms of pay and the lack of transparency. It is pretty clear that there is massive inequality in bargaining power at this moment. Are there any plans to address that through legislation?

That is a really important question. Of course, where there are concerns in workplaces around the use of technology, we want to understand those and build the evidence base to understand how best Government can respond and rise to this challenge. The consultation that I mentioned earlier to the Chair around workplace monitoring technologies, which we have launched today, is building that evidence base around how digital tools are used by employers to track workers, collect and analyse data on them or make decisions based on information about workers and their activities. We can build that evidence base to make sure that we understand what is happening on the ground in workplaces, based on evidence of current usage; how current and existing requirements are being applied, so the existing responsibilities of employers under the current legislation; and how workers are or are not kept informed about and engaged with those decisions. We have a range of options in the consultation, from regulation to statutory guidance, to understand what is happening and what Government now need to do to respond. It is open for a number of weeks.

This covers pay as well as general monitoring in the sense that we would understand it.

From my perspective, good adaption and use of technology is driving up good-quality work, and pay is essential to good-quality work and experience. If workers and their trade unions, where they are recognised, are therefore not being engaged in the deployment and adaptation of technology in workplaces, we ultimately do not get those good-quality jobs, which pay is key to.

You have written to us while we were in session. Thank you very much.

Yes. We are very happy to keep you updated throughout.

Unfortunately, we received the letter at 3.55 pm, which was too late for us to study it before the session, but we look forward to reading it in detail.

I will kick off on that, and others may have things to add. You see different levels of intervention in the skill space. I know that Skills England was here earlier. We were having the conversation earlier about its—

It assured us that what was needed was not in place at the moment.

Yes, but you see progress from it.

There is always progress.

Well, yes. The AI practitioner apprenticeship, for example, is a really interesting example of a skills intervention. Because you are teaching people the human interaction with the AI, rather than how to use a particular model, the skill that you are teaching them has the potential to be sustained. We see the work with AI Skills Boost and the drive to upskill 10 million people by 2030. If we are successful with that and it resulted in a 1% productivity gain for 10 million people, you would be looking at about £7.5 billion added to the economy just from that.

One of the things that we are working on at the moment as a test case is the idea of a tech town.

You can join the queue. By bringing together not just all the Government support but all the different support that is available from providers, we can learn lessons from that in terms of how to support high street businesses, small manufacturers and the interaction locally between businesses and the public sector. We are trying to develop a model.

Is this a radical devolutionary proposal that is currently being developed or is it a policy offer that is on the table already?

We are working very closely, but it is less a radical devolutionary thing than it is an experiment.

It is Barnsley—yes, exactly. We are working with people in Barnsley to really pile in all the resource that is there from both the public and the private sector to see what can be achieved and to learn from that.

We are leading it along with DSIT.

Kate, you have a track record of working to protect working people and have done a lot of work in your current role to enhance people’s rights at work, which we are very grateful for. You will probably realise that there is still a lot of scepticism and concern about AI and how it will impact people’s jobs. Without scaring people, I presume that in the Department you have done some work about how, if we do not get it right on retraining and on lowering and closing the skills gap, jobs could be lost. What work have you done to factor that into the work that you are doing now and to make sure that we learn the lessons, before the lessons potentially happen, to keep people feeling secure in their jobs?

That comes back to the wider approach and what this Government are hoping to achieve with our Make Work Pay agenda. We fundamentally believe that you cannot grow the economy with people in insecure work. That applies to technology, contracts and so much more that we are working on in the Department. When we look at AI, in particular, and the impacts that we are seeing, I can completely understand that when people see BBC News notifications about research reports saying that x number of jobs will be lost, they ask, “What does it mean for me in my job and my role in the future? How can I best prepare my career and my skill base to adapt and respond to that? How will my employer support me?” When we look at the evidence today specifically around AI reducing employment, at the moment there is quite inconclusive evidence about whether AI equals a significant reduction in employment. There are a lot of wider instances where an employer might make certain decisions about their workforce, but where AI is a particular factor some employers might say, “Yes, we are introducing different types of technologies to increase efficiency or change how we do things.” There is only one way to get that right, as I keep talking about. We want businesses to make these changes because we know that, if done well, they can create a more productive workplace and those businesses can grow and compete internationally. The people who ultimately will be using the technology are the workforce. You want it to work for them. You want them to feel it is fair, the process is transparent and it will ultimately deliver its objectives, which I would hope the business wants to achieve. While the evidence at the moment is quite inconclusive about the potential impacts, if we do not get it right, the impact will be on people’s quality of work and future.

Just to follow up on that, are there any particular sectors or regions that you have concerns about right now?

In terms of particular sectors, DSIT’s assessment in January touched on this. Kanishka can talk about the evidence base that DSIT is building and some of the estimates from the IMF, and I will let him come in on DSIT’s work and the stats on this. When you look at AI’s ability to displace jobs, patterns and sectors, it is quite mixed. If you look at the past couple of years, particular sectors have been impacted by lots of changes in the labour market and by wider macroeconomic issues rather than by AI itself. I will let Kanishka come in on the particulars.

The situation that we are in is exactly as Kate says. There are good reasons for people to be worried, based on the micro evidence, which is to say that the scale and speed of AI as an economic force has been very unusual. In 2023, AI as a sector took something like 180 days to add $1 billion of revenue. Now, AI as a sector is taking less than two days to add $1 billion of revenue. The scale of that uplift is exceptional. The second micro evidence is that, when people have done theoretical work to understand the exposure of tasks to AI, they have found that models’ capabilities are increasing so much, starting with language, to coding, to data insights and increasingly to mathematical calculations, that the scope of possible exposure theoretically is very high. That is the micro evidence. Against that, you have the macro evidence, which, as Kate says, effectively shows, at least in the UK, no signs of net job loss and no systematic patterns of bias in terms of age, gender or geography. What are theoretical reasons, then, to believe that we have some view of what the answer to your question might be? This feels very unlike some past technological advances, in that, if you just take those task exposure measures, previous technological revolutions were, on a relative basis, better for higher-paid workers, for knowledge-intensive workers and for women, and they were better in places of agglomeration, often cities. Again, it is very important to say that we do not have good empirical evidence on these impacts. Theoretically, if you look at task exposure to AI, then highly paid jobs, knowledge sector, female heavy and white collar heavy, from a task bundle point of view, are the ones that appear to be most exposed.

We have a hard stop in about three minutes.

I just want to follow up on that. I understand that the AI and future of work unit is the body looking at this. Unfortunately, apart from a press release in January, there has not been a great deal of information about what the unit has been doing. How many people are working in the unit? How often is it meeting? What is the ministerial accountability? When can we see some outputs?

The AI and future of work unit was a cross-Government unit that we set up. We then, as a result of the Chancellor’s Mais lecture, upscaled it and broadened the scope, and it became the AI Economics Institute. We appointed a chair, Simon Johnson, a Nobel prize winner from MIT, who has looked at the impact of technology on labour markets. We made a first significant publication a few weeks ago at London Tech Week, working with private data providers to look at a real-time understanding of the impact of AI on jobs, which is the foundation of my prior answer to Chris on the empirical position. We are recruiting right now. We continue to scale up that organisation. I am very happy to write to the Committee with the latest headcount, but certainly there are staff in DSIT and the Treasury who are working on it.

You each have a minute to sum up your advice to the next Prime Minister about how we harness AI for good.

This will be my advice to the Prime Minister, but it is also something that should not be lost from this conversation, particularly the conversation about productivity. Part of the reason that we have a productivity problem in this country is that we have poor dynamism within the market. AI is a huge opportunity for smaller companies to challenge complacent larger companies. Earlier on, eBay told me a story about a woman who was trading who built agents around her one-man-band company.

We heard the story from the chair of Skills England.

He was in the same meeting as I was.

It is a very interesting story.

That is the opportunity that we need in order to build scale across the economy.

Therefore, what is the one big policy recommendation you would make?

We should put our shoulder behind tech adoption across the economy because traditionally we are pretty poor at that in the UK.

We need to ensure the successful adoption of tech across workplaces in the country, and growing adoption and uptake from businesses, in order to grow our economy and raise living standards. If we do this right, there is a real potential for workers all across the country to benefit in terms of living standards. For me, it is about that worker voice. They have to be involved in deployment, shaping what that technology looks like. We want to build an evidence base to understand how we can respond and do that best. That is why the work that we are doing at the moment is so important. We really look forward to working with the Committee and keeping you updated, particularly on that consultation and the wider work about how we get that right and deliver.

Whatever your objectives are, AI is going to be a central driver. If you want to reindustrialise this country, the most private capital-partnering way of doing that is to build chips in this country so that you can capture the economic opportunity of our lifetime. If you want to make sure that you are spreading opportunity, you do so with clusters right across this country that have deep strengths in AI-enabled sectors. If you want to make sure that you are building a sense of national security and sovereignty, the central question in whether we do that or not is the question of our AI capability. At the heart of it, if you want a true Labour vision of state capacity driving that, there is no better example than the Sovereign AI Fund, the AI Economics Institute, the AI Security Institute and ARIA collectively having built world-leading positions, compared to the very best of the private sector, at the heart of Government.

Time is up. Thank you very much indeed. That concludes this panel and this session.