Business and Trade Committee
Oral evidence: Artificial Intelligence, business and the future of the workforce, HC 125
Tuesday 16 June 2026
Ordered by the House of Commons to be published on 16 June 2026.
Members present: Liam Byrne (Chair); Dan Aldridge; Antonia Bance; Chris Bloore; John Cooper; Sarah Edwards; Alison Griffiths; Leigh Ingham; Justin Madders; Charlie Maynard; Mr Joshua Reynolds.
Questions 114 - 126
Witnesses
II: Vinous Ali, Deputy Executive Director, Start Up Coalition; Paul Wilson, Director of Policy, Federation of Small Businesses; David Spence, Head of Applied AI, Innovate UK; Dave Sellick, Founder and Director, Sidgrove.
Witnesses: Vinous Ali, Paul Wilson, David Spence and Dave Sellick.
Chair: Thank you very much indeed to our witnesses for coming along. There may be a vote at any moment, in which case I will suspend and we will return as quickly as possible. Leigh Ingham is going to open the questioning.
Q114 Leigh Ingham: It would be interesting to understand how widely SMEs are currently adopting AI and which tools you are finding are the most common. Vinous, could I start with you please?
Vinous Ali: Hello. I am from the Startup Coalition; we are the policy voice for start-ups and scale-ups in the UK. We are building the tools that I hope the economy at large will adopt.
What we are seeing in terms of adoption is patchy; that has been the story of UK digital adoption more broadly, historically. It has taken us a long time to get to 70% penetration levels for, say, cloud technologies. We are seeing those that are adopting AI much faster and then there is a long tail that is far slower to come on stream.
The types of tools are really diverse. In some sectors you see AI being used in a really clinical, surgical manner—tools that have been developed for niche purposes, some of which our start-ups are building. It is probably a safe bet to say that where we are seeing the greatest level of adoption is where we are seeing the generic LLMs come on stream. We know that at the moment adoption rates among the British population at large is running at about 50%, which is actually really strong compared with other nations. I would hope to see that translate into more commercial settings, but we have seen this divide before. For example, we all have one of these mobile phones in our pockets and make great use of it and then we go into the office and get out our fax machines and so on. Hopefully we can leapfrog that this time round, but there has always been a historical split. I would say that where AI can differ is that it is a much more instinctive technology and there is a real opportunity to drive and accelerate adoption much faster in a lot of scenarios, particularly where it is embedded into key systems and workflows.
Q115 Leigh Ingham: In our last session we heard that 62% of workers in large organisations are using AI, but may not have been trained in it. Do you have any similar statistics across the sectors you work across?
Vinous Ali: I do not, and that is partly because in the start-up community we are operating right at the frontier. These are folks who are dedicating their livelihoods to building and innovating at the frontier, so they are AI and digital natives from the get-go really.
Q116 Chair: Mr Wilson, what is the general story?
Paul Wilson: We have seen a huge increase in AI usage among small businesses over the last couple of years. Up to 55% of small firms are consciously using AI within their business and on average that is 4.7 different uses within the business. Obviously it is variable between sectors; information and communication, professional and scientific are the biggest users. What we see is newer businesses—businesses formed in the last four years—are 90% likely to be using AI. For obvious reasons, they can adopt it almost from the get-go.
In terms of uses, I would say the main thing is to improve your operational efficiency; some 60% are using it to draft written content; a reasonable proportion are using it to generate and edit images or videos, things such as that. I actually believe it is being underused in some contexts. For example, in public procurement bidding, which seems such a really obvious thing to use it for, it is still only one in five firms submitting bids, there is a big opportunity there to use it a lot more. Some are using it to a real depth; around one in five are using it to develop new products and services or to make forecasts and 4% are using bespoke AI tools within their business.
David Spence: I work for Innovate UK, the UK’s innovation agency. From our point of view of the companies we have worked with and will work with in the future, the feeling is that adoption is accelerating, but it is very uneven. It is uneven across different sectors and between different sizes of organisations and there remains a large gap between that experimentation with AI at that level of trying productivity gains, that sort of thing, and of embedded scaled adoption.
Q117 Leigh Ingham: Thank you very much. David, what is happening in your organisation?
Dave Sellick: I am a little different; I run an accounting firm in the SME space, working with around 50 various sized businesses up to a £5 million turnover. I work across all areas of business from a finance perspective, not just the compliance side. I am on the frontline, so I will give you the story based on what I am seeing, rather than just the numbers; I will talk about what is actually happening at the cutting edge.
As an accountant, I experiment extensively with technology and am probably one of the highest adopters of AI—or gen AI, which I presume we are predominantly talking about today—in our industry. However, I am also one of its biggest sceptics. In terms of my clients, Vinous is probably right; adoption is indeed flaky. When I ask how they use AI, there is often silence before they say, “Well I’m using a chat interface,” or give you a few different examples. There is no overwhelming sense of, “This technology is completely transforming our business and we need more support,” or, “It’s the best thing to ever happen to us.”
Just to talk about the key areas I see them using it, as we have not really talked much about the different ways, you have the Copilot style of using Generative AI, using something such as ChatGPT, Claude Chat, or the Copilot chat interface. Most of my clients are doing this. They use these chat interfaces, effectively as a very clever—I will say very clever, not intelligent, because I do not think gen AI is actually intelligent—co-worker that sits next to them, that they can tap into for anything and everything all day through their work.
If you were to ask me what ways they are using it, I would say that is a bit like asking what sentences my clients write with their keyboard: it is everything, and it is contextual to how their businesses work. As mentioned earlier, some key areas would be email creation, meeting notes transcription, business research and marketing content and branding creation. In some instances, I have seen these kinds of workflows result in clients talking about this potentially either replacing employees or meaning they may not have to recruit further employees.
The next way they are using it—I advocate both the first and this second way—is using generative AI to do what we refer to as vibe coding. Using the technology to send off agents to create code, to build assets, to build tools, to build websites, to build slide decks—tools that might not necessarily run with AI but can be created with this technology. I see this as a highly effective way to work with my clients, and an area that should be supported.
The last area I want to cover is using generative AI as a pilot, not a co-pilot. By that I mean the execution of work from end to end. This can be seen as a double-edged sword and actually a significant risk in many areas and we will come on to that maybe a bit later.
Chair: We will stop you there just because we are going to get into benefits with Sarah Edwards.
Q118 Sarah Edwards: It was mentioned earlier that the adoption was uneven, which is quite key to benefits, because we have PwC estimating that GDP could go up by 10% in the next three, four years. The British Chamber of Commerce is citing SMEs as being the most optimistic in their adoption of AI, but at the same time an SME is 0 to 249 people. It is not that helpful—although we have some interesting breakdown statistics—so I wondered if, perhaps, you could each focus on a different size and scale of a business or a particular sector in the answers, as to where you think there is the greatest benefits for this use? We have heard a bit about finance being quite obvious and we are talking about manufacturing as a potential that is probably quite difficult. Do you have some other examples?
David, I do not know if you want to finish your point, then I will move along the panel because you might be able to talk then about the benefits that you see from the accounting profession or from your specific clients and what size they might be?
Dave Sellick: I will start from where I left off in terms of where a lot of people will see the most significant benefit of AI, or the area where this almost doomsday narrative is being created that it is so good that it is going to replace people or even professions. That comes from what I was going to talk about there, which is using AI to effectively run end-to-end process.
I personally do not see this as the biggest opportunity, but you will hear from a lot of people that they do. They will talk about professions such as accounting, potentially legal, and coding. These kinds of industries are very much in the limelight in terms of the potential to use generative AI to almost replace entire professions. Of course, I may be speaking here from the point of view that I do not want that to happen to my profession, but as an overall within the economy it is perhaps a beneficial thing for these professions to be completely automated.
From my perspective and what I am actually seeing, I do not think that is the most effective use of generative AI. It is what we refer to as a probabilistic technology; that means that it is probably correct when it generates output. It does not mean it is correct. It creates plausible answers—it does not create correct answers. Therefore, running this technology autonomously—not just generative AI but any AI, or in fact any technology—means that you will have a level of this probabilistic decision making happening. For mission-critical jobs and outputs such as finance, potentially legal, and in some instances code as well, when we come to security and so on, we really need to consider the risks as well. I know no one has highlighted that yet.
If we talk about accounting, risk can also extend to things such as tax revenue to the HMRC, with clients choosing to use this technology to tell it to reduce their tax and create this riddle that needs to be deconstructed. Where I think it is actually effective are the areas that I just mentioned: the clients that are using this technology as an opportunity to augment human skillset and build tools that run what we refer to as deterministically, not putting risk on the table but breaking down barriers that otherwise would have existed in client areas such as coding—
Q119 Chair: May I just speed us up, as we have a vote in a minute? Mr Spence, would you agree with that analysis? What else do you see going on?
David Spence: I would agree with that entirely. What we are seeing at the minute, the different benefits, the different productivity gains and all, it is exactly like that; no doubt about it.
One key thing I mentioned previously, these being used—some can be off-the-shelf services—but it is not just that: it is a question whether it is really being embedded into core AI activities within the company. That is probably harder for a smaller company. A DSIT survey found that 75% of businesses using AI reported a 50% increase in workforce productivity. It is definitely working there, it is definitely involved, but there are so many other things to come. We work with a lot of companies using AI to determine crop safety, that type of thing. There is so much more that we are not actually seeing in the markets yet.
Q120 Sarah Edwards: Paul, across the range of businesses that you see, do you have specific examples where you can say, “This business in particular at this size really found that this and this worked really well”? How are they approaching that, and where can we latch on to the real positives of what they can do? I am just trying to bring it back to an example, because we are talking at quite a high level otherwise.
Paul Wilson: Relatively unusually, we see fairly consistent productivity and revenue gains across all sizes of businesses. I know that is a super unhelpful thing, but AI is very accessible for the self-employed and micros, as it is for larger businesses. We see 11% average productivity gain and 3% average revenue gain. The thing that is quite interesting is, if you take certain steps in how you adopt AI and you engage your staff in how to adopt AI, you are much more likely to be one of the businesses that sees a significant revenue gain. If you have a policy on how you are going to adopt it, you are much more likely to see that because you are doing it more confidently, within parameters that you have set.
In terms of what they are doing, the best examples are businesses that are freeing up their own time to do more of the thing that adds value to the client and, frankly, that they enjoy—for example, the plumber travelling between jobs using AI to take notes to produce quotes so that they do not have to waste time doing that separately. Businesses that are using AI to record the client’s needs and produce an initial draft proposal for working with that client. We have a brand consultant who actually has an AI version of themselves; that is the first interface that the client has. They know it is AI, and he can then spend more time getting into the really knotty stuff once the AI has already diagnosed the easy stuff. These are some good examples of basically freeing up time and improving the quality of the business owner’s life, and hopefully that of their staff as well.
Q121 Sarah Edwards: Vinous, start-ups start out straight away and quite a lot of your members are probably in that zero to four-year category I would imagine where they are already going, “What are all the tools I can possibly have that make me really efficient?” Perhaps you have some great examples of how they have gone exponential growth and adoption. You can name drop as well, feel free to do that.
Vinous Ali: Exponential growth is definitely the business that I am in. What I am really actually most excited about though, Sarah, is just the fact that AI tools are lowering the barrier to entry in such a remarkable way. I really believe that should help spur on entrepreneurship if we can get the right sort of safety nets and guardrails around that. A few years ago, if I had an idea, maybe I did not really know how to take it. Now I can spend a weekend and essentially create an app, create a marketing strategy, understand what the legal frameworks are that I should think about before I go forward.
What we are seeing from the founder end of things is actually that AI is acting as an accelerant to getting their idea off the ground. Then you can go into Innovate UK, UKRI, the brilliant funding programmes that are out there right across the UK, the incubators, the accelerators, to take that idea from zero to one. For me, that is the huge opportunity. I know in the last session you talked about young people being digitally native, excited about this tech and so on. Given the NEET crisis, there is something there about how you can use the technology to really accelerate their ability, to sort of test their own ideas and be their own boss. That is what I am particularly excited about.
Q122 Dan Aldridge: The growth potential and lowering the barrier for entry was mentioned earlier. As constituency MP for Weston-super-Mare, which I would imagine has lower levels of AI adoption than others, it strikes me that the risk is that we leave communities behind. What is the role for Government in that and what is the role for organisations such as yourselves? We have some members of Startup Coalition in Weston-super-Mare, but also FSB and others, so it would be useful to get a sense of those barriers and how they are manifesting in communities.
Vinous Ali: Where Government can be really powerful here is adoption of AI within the public sector. I have a thesis—I may be wrong, but I hope I am not—that actually you could see a faster acceleration and diffusion of AI in the public sector because it is a much more structured employer. For example, the NHS has clear L&D budgets and so on, so you could actually hold people along that journey with you. Whereas in the private sector you have often seen L&D budgets decimated over the last few years. Government adopting this technology themselves and rolling it out in a way that is responsible, that is very much human and in the loop, can start ensuring that there is confidence and trust built in communities. That is going to be critical to adoption.
At the moment what we are seeing is that, despite the fact that 50% of Brits are using AI already and the fact that—we have some polling coming out next week, and my comms colleague is probably going to kill me for trailing it—65% of Brits believe that a strong economy is only possible with a strong tech sector, on the other hand public attitudes to AI as a whole are taking a bit of a nosedive. The only way to rectify that is if people start interacting with it in their daily lives. Whether that is in the jobcentre, in the NHS, in schools, and so on. Government have a strong signalling role to play that this is a technology that we should be harnessing, that there are benefits there, and then that is where we can sort of get that flywheel of diffusion going.
Paul Wilson: There are definitely some sectors, such as hospitality, where it is harder to see obvious use cases for AI. Therefore I worry a little about high street small businesses, particularly within the hospitality space. Then you have to look at what we can all do to try to encourage those businesses. It is not bricks versus clicks—there should be bricks and clicks businesses: how do you take them online? How do you get them to use AI for their booking systems, for their social media, where in fact it can be really powerful and give them that extra resilience so that they do not get left behind?
Q123 Antonia Bance: Paul, I will start with you. How is AI affecting jobs skills and ways of working in SMEs, and how are businesses building the capability to use these tools? We have spoken a little about this, but particularly about the skills gaps: where are they most acute, and what more needs to be done on making sure people have the skills that they need? Paul, I will start with you.
Paul Wilson: I will return to the point I made earlier: it is a good news story that the businesses that have engaged their staff well in how they roll out AI—as with any tech project—have seen the best results and have improved the quality of jobs for their staff as well. We want to find ways to encourage more businesses to do that. That is something that FSB can look to do, but also Government can look to do with their business support offer.
In terms of skills, skills gaps and knowledge gaps are still a barrier; they are less of one than they were two years ago and you can see that in the broader extent to which businesses have rolled out AI. One that I would pick up on though is there are widespread concerns—27% of small businesses have concerns—about how to implement AI securely, ethically and responsibly. We heard the previous panel talk about that. That is holding them back from confident adoption, by which I mean that they could be using it more, they could be using it in other bits of the business, they could be monetising it better, but they do not.
I will give the example of a fashion business that looks to incorporate AI into t-shirt design, but then it steps back because it realises, “I don’t know who actually owns the IP around the design on the t-shirt. Is it the client who has put the prompt? Is it the provider of the AI model? Is it me the business?” Helping more businesses to navigate that and come up with their own policies is important.
There is of course a huge role for schools, colleges and universities here. It would be great if we can think about how to build AI into project work and coursework and then have people coming out of the education system that are already used to using it. I would say that DSIT has really good ambitions in terms of its skills hub, its skills boost—which we are proud to be a partner of—looking to help 10 million individuals and small businesses. The ambition levels are good; we just need to build upon that.
Chair: The Minister is on his feet, so I am going to come to Justin Madders very quickly then, if the bell has not gone, I am going to come back to Alison Griffiths. I just want to zero in on your short advice for Government.
Q124 Justin Madders: AI adoption schemes for SMEs: what has worked in the past and what do you think Government should be doing and funding?
Chair: David Spence, do you want to tell us why everything is okay?
David Spence: Do you mean on skills specifically?
Chair: No, general diffusion of AI through small and medium-sized enterprises.
David Spence: Obviously I am representing the role of government in this situation on what we can try to do. For me, at a high level, it is really removing the risk, or at least reducing the risk, for SMEs to adopt AI. That could be a number of things to give them that confidence in adoption: skills, data, suppliers and assurance. When we say skills, it is not necessarily just about the skill level of each individual being able to do their job, but a skills level targeted to the right people, so that the CEO team, the C-suite team, know exactly what they are trying to do and how to implement it, right the way down to the people actually doing the work. Government should be a good facilitator, a convener of experts, bringing the right people together so that those suppliers, those people with the solutions, can work with the people who are looking to adopt AI. In the middle of that somewhere is that level of assurance that those adopters know that the tool that they are looking at has been assured and can be safely adopted within their organisation.
Vinous Ali: One area of interest to us is looking at how you can incentivise the adoption of AI through tax credits and basically make it the problem of accountants—sorry, Dave. But essentially, we have an R&D system that perhaps is not as flexible as it could be and so looking at things such as whether it could be used for compute or for other areas within the AI space to accelerate the adoption of AI is one really interesting area.
Another one is whether you can build the relationships between large corporates—apologies, not SMEs—that have venture arms and the domestic market. We know that at the moment corporate venture capital does not really serve the domestic market. Can Government, through an accord similar to Mansion House, for example, create those bridges between the start-ups that we serve at Startup Coalition and those sorts of large corporates that are funnelling money into the system, to get that exchange of knowledge and ideas into a more formalised manner?
Paul Wilson: There is a huge opportunity here—we have calculated a £42 billion opportunity—if you can get the businesses that are using AI to use it half as much and half the businesses that are not using it to use it. There is massive economic opportunity and that is just the revenue growth benefits. To pull out a few things, we agree on the tax relief side of things—such as R&D tax relief—and expanding that to support AI adoption could be really powerful and then it would be communicated through accountants.
There is a role for Government to regulate upstream large AI models. There is a lack of trust at the moment; there is a belief that the models produce inaccurate responses—and that belief is stronger in those using them than those not, which I find quite interesting. To help businesses implement that confidently, Government could look to roll out something called a responsible AI essentials programme—something that is built along the line of cyber essentials that helps them define the tram lines within which they are going to use it, maybe get insurance and then move forward with confidence on it.
Dave Sellick: Obviously, we are talking about gen AI here, assuming that it has a strong positive ROI—that is, that there is a correlation between the amount of AI that we roll out and that businesses will improve, revenue will improve, bottom line will improve.
I am not seeing a quantifiable ROI on the front line yet, so I am just saying there is still speculation around false end on gen AI. Perhaps what is a more reliable bet is that in order for gen AI to be used effectively—I will not go into the specifics because it might get more technical—you need to have digitalisation as a whole. For example, without going into the details of technically why, if my client’s data is not in the cloud, accessible, well structured, accurate, they really cannot leverage a huge amount of the power of AI or gen AI.
I will not give the percentages because I have read different percentages, but I know that in terms of digitalisation, the statistics are still probably surprisingly low, particularly in my industry. If we could get all small businesses on to a cloud accounting technology before we start pushing generative AI, then that would have a much more significant benefit. I would use the analogy of pushing gen AI without digitalising small businesses is like ordering more icing without cakes to put on.
Q125 Charlie Maynard: I am going to be a bit contrary and negative maybe. Both panels have been very positive about a lot and there is obviously a huge amount to be happy about. But we have China and the US going faster and further than the UK is. We have large multinationals often going farther and faster than smaller companies. Now there are some winners and some losers—I get that. We have £400 billion being spent by the top five AI companies globally. That is going to take a lot of general admin cost and jobs out of the market almost certainly, which is going to lead to quite a lot—maybe a massive amount—of dislocation and that could happen very quickly. Is that something that is just nonsense? Or is that something that we do not want to recognise and just carry on? Or do you think that is actually a real risk that we should be addressing much more seriously? How worried are you about that sort of stuff?
Vinous Ali: I am happy to go first on this one. It is really important that we the UK pick our races and we know where we should compete and where we should not. There has been a lot of chatter recently after the Anthropic story about, “Okay, we need to go full throttle on building full stack sovereign capability in the UK.” As you said, the billions that are being pummelled into doing that by some of largest players, the US and China, plus, as we heard from the last panel, our own energy constraints and regulatory hindrances—whether that is copyright or other things—mean that, actually, developing our own sovereign LLM is possibly not going to be where we win the race.
Where we are incredibly competitive is in areas of historical strength, whether it is financial services, life sciences and so on, and at the application layer of the stack. There we should really be doubling down and it is great to see from Government things such as the Sov AI unit getting up and running, giving our start-ups both compute power and direct investment. But to your point, Charlie, it is about understanding the global landscape, making sure that we remain open while ensuring that we have leverage and influence, given the geopolitics of today, and picking the right races.
In terms of that point on dislocation, if we get that first piece right, we can support that transition into whatever the new economy looks like. I certainly do not have a crystal ball, so I will not speculate there. It is about making the right policy choices today, based on the landscape as we see it, not as we would like to see it, that can ensure that we seize on the upside while minimising on the risks.
Paul Wilson: It is important that Government can move really quickly if we see any of these risks accelerating. One of the things we would recommend is that the digital regulation cooperation forum, currently an informal grouping of regulators, be put on a statutory footing and therefore given responsibility for AI regulation and be able to react more quickly. Our evidence does not show that businesses are using AI to displace jobs at any great rate—in fact, where they have had to cut jobs for cost reasons, they are using AI to kind of come in and fill the gap—that is something that Government should monitor incredibly closely. The AI Economics Institute and bodies such as the Low Pay Commission should be looking at that interaction between AI, employment costs, jobs and things and, again, be prepared to act quickly if it looks like things are changing. But as I say, we are not seeing that risk significantly at the moment.
David Spence: I do not have a great deal to add, to be honest. Vinous put it very well about picking your battles and looking for those unicorns that the United Kingdom has expertise in, can develop and can become world leaders. In terms of the regulation, from my point of view the general feeling at the minute is the UK has a good balance of regulation that is not too risky and does not stifle innovation. From that point of view, I have nothing to add.
Dave Sellick: In terms of whether the AI is going to take people’s jobs, obviously my industry and profession are the target of a lot of that conversation; I will not speculative or say fully maybe what I think on a speculative basis; I will just say what I see in terms of evidence. I believe the statement that it is going to take people’s jobs is exactly that: speculation. I do not see any evidence that it can happen at the moment based on the mechanics of how this technology works—it is probabilistic technology that cannot automate what we do as accountants, regardless of what you might hear in the narrative.
If I am honest with you, I do this job every single day and I will tell you that, on the basis of intuition. I am looking at what this technology is doing; I am using the frontier models for 12 hours a day in every way possible and imaginable. I am not seeing any evidence that this technology can replace what we do in a consistent, accurate, high integrity way without leaving our clients potentially in positions where, for example, they may be subject to a liability or a fine, or leaving ourselves in a liable position. We have seen evidence of this already with accountancy firms, in terms of things that have been put out.
My answer is no in that regard, but that may well change; if it changes then, yes, I expect there still to be some level of disruption in our industry. We are obviously seeing that across small businesses, and accounting is one of those where jobs have been displaced. I will not comment on whether that is actually because of gen AI, but I am sceptical that it is because of gen AI. In multiple sectors we are now starting to see re-employment of those people who have been displaced. I also want to draw your attention to—
Chair: I am going to sneak in Alison Griffiths’ question at this point.
Q126 Alison Griffiths: To what extent is SME AI adoption dependent on large technology providers and what does that mean for cost, choice and building in-house capability? I am going to give you part two at the same time so you can answer both. Are certain vendors better for SME AI adoption and what is different between them?
Vinous Ali: The only thing that I will say here is that the events of the last few days have shown that relying on one model provider is a supply chain risk. What we really want to see is a diversity and a flourishing of providers being used. Certainly different models have different capabilities and you might want to use a different model based on where you are deploying it. For example, if you are working in a highly regulated environment or a secure environment—that goes for the MOD as well as the private sector—you might want to use a model that you are able to secure and take offline so that, essentially, it cannot be switched off. But more broadly speaking, it is about having that sort of mix.
Personally, I want to see the UK remain open, because that is the way we remain competitive. Whether that is proprietary models or open-source models, there is a role for all those. It is about getting that mix right and ensuring that we recognise that, like anything else, if we are overly reliant on one provider, then there is a risk.
I just want to come back to this issue of public procurement though; I will be super quick. What we see is that UK Government actually prefer large vendors, partly because there is this risk aversion to buying from our start-ups and scale-ups. The old adage goes that no one was ever fired for buying IBM, and you could replace that with a number of others. Again, can the public sector lead the way and make sure that we are really realising the benefits of the wonderful start-ups that we have in the UK today?
David Spence: There probably is a growing reliance on a small number of foundation models, cloud and AI tools. The effect that it has had is it has accelerated development, it has accelerated adoption, but it creates a resilience issue.
Paul Wilson: I agree with the points that Vinous and David made. A lot of small businesses will get this stuff on subscription, so you have to look at the ease of exiting those subscription contracts and the ability to switch and the interoperability of systems, both AI and non-AI. Those are all important factors. I guess one of the upsides of having a relatively small number of large players in this space is that it makes it slightly easier to regulate those large players and provide that transparency over what those big models can do. There is a bit of opportunity there.
Dave Sellick: It is an issue. I read an article from a journalist today who had apparently seen the audits of financials for OpenAI, and it suggests that their losses were over $100 million a day in 2025. When that is happening, there is of course a risk that those kinds of businesses potentially may not last. If we have embedded gen AI—in particular I mentioned the autonomous type of generative AI—into the operations of our small businesses, or any business, quite frankly, and that model goes down, then yes, based on those figures, it could potentially fail.
There is a resilience issue there. That speaks not just to the fact that we need to look at supporting businesses to look at multiple models and being agile to change, but to being wary and potentially working with businesses to disclose the risks that happen when we are embedding this technology autonomously. We are very early at the moment; if people indulge too much in a narrative and they go fully in on this technology, we still do not know what the next month, two months, three months, six months will look like for these models and which ultimately will be the frontier models or the models that survive through that period.
Alison Griffiths: You have just timed it to perfection if I may say so. I just glanced at the screen and we are about to hear the Division bells ring. Thank you.
Chair: Thank you very much indeed to our witnesses. If there is anything that struck you over the course of what you have heard today, then please drop us a line by way of further evidence, we would be very grateful for that.