On social democracy in the age of increasingly intelligent machines
In these pages in 2022, David Klemperer and Morgan Jones authored a piece about a group called Labour for the Long Term, an attempted entry of Effective Altruists (EAs) into the UK Labour Party. I remember reading this with the sense that it had crystallised something for me that I'd previously felt at more of an aesthetic, almost instinctive level: 'I'm not sure these people belong in a progressive political movement'. If that seems an odd disposition to have towards a group of people who pledge to give away some/most/all of their salaries to the charities they think will be most impactful, often those seeking to combat malaria, then it's worth examining some of their moral intuitions a bit more closely. Longtermism, by placing infinite value on future happiness, can quickly end up justifying things in the here-and-now that most people would find abhorrent. Klemperer and Jones argued that for all their talk of doing the most good, EAs lacked an analysis of power. This was made to look rather prescient when, just months after the piece, EA poster boy Will MacAskill found himself inside the blast radius as Sam Bankman-Fried's scam crypto empire collapsed; MacAskill had clearly rationalised this association through his thoughts of the good his money could do now — but, more importantly, in the far-off future. This sordid affair made plain the pitfalls of a politics that not only tolerates extreme wealth, but depends on it - and is happy to prostrate itself before it.
The effective altruist community was throughout the 2010s talking about 'existential risks' writ large. As well as artificial intelligence, they were concerned with advocating for things like new biosecurity institutions to prevent pandemics, animal suffering in factory farms, preventing nuclear annihilation, and even the odd mention of climate change. If you're thinking, 'you don't hear much about longtermism” anymore', it's because the people involved, more or less, have ended up in the AI safety world.
The Labour for the Long Term website is now defunct. Haydn Belfield, who was on the advisory board of Labour for the Long Term, now works as a research scientist on ‘Frontier Planning’ at GoogleDeepMind, the company’s AI lab subsidiary.. David Lawrence, the founder of Labour for the Long Term, briefly a Labour parliamentary candidate in 2024, went on to become, among other things, a policy consultant to the AI safety group Control AI, according to his LinkedIn profile. Last month The Daily Telegraph reported Wes Streeting accepted £37,500 in donations from Labour for the Long Term over 2022 and 2023, despite what it claims were significant concerns among some senior figures in the party over the source of the funds; the Telegraph also revealed that one month before the first donation to Streeting, Lawrence had accepted a £500,000 gift from the now convicted fraudster Bankman-Fried. Lawrence has since founded the think tank the Centre for British Progress, which describes itself as 'producing concrete ideas for an era of British growth and progress', including a stated focus on 'building AI infrastructure and improving AI adoption'. I cite the paths taken by Belfield and Lawrence not necessarily to impugn them, but to bring them into focus as a microcosm of the narrowing of the EA horizons from a wider focus on 'longtermism' into being principally concerned with the world of AI safety. Many more in Silicon Valley have been on similar journeys.
Naturally, then, when discussions of Artificial Intelligence began to gain a degree of mainstream interest with the launch of ChatGPT in 2022, I was extremely well disposed to believe this was yet another scam, with lots of the same cast of characters at the helm. For the next three years I would have flirted with all the AI-sceptic arguments readers will be familiar with and may still believe: AI, like crypto, has no underlying value or utility, with Altman as the new Bankman-Fried; Large Language Models were nothing more than 'stochastic parrots', a metaphor to disparage them as statistical systems that mimic human language without any true capacity for understanding; any talk of 'AI safety' involved one buying into the framing that US Big Tech wanted, of themselves as stewards of this incredibly powerful technology. That mythology was one they needed to maintain in order to justify vast debt-driven capital expenditure on data-centre infrastructure, and to justify their valuations to prospective shareholders ahead of their IPOs — all of it driven by a Venture Capital model with hype as its main currency. They were trying to 'fake it till they made it', and, I believed, like Elizabeth Holmes and Bankman-Fried before them, they weren't going to make it. In short, they needed the safety ruse to stop the bubble from bursting. Of course, some of these arguments have elements of truth. But I have derived little pleasure from my intellectual journey towards realising that, when it comes to the trajectory this technology is on, lots of those making these critical arguments have been consistently, hopelessly wrong, and that the twenty-five-year-old Silicon Valley AI safety nerds making $800,000 a year have been fairly dependably, annoyingly, right.
A transformative general-purpose technology
To take just one example, Paul Christiano, one of the world's leading AI safety researchers, has spent years making predictions that have proved remarkably prescient. Back in 2018, against those forecasting a sudden, discontinuous 'intelligence explosion', he argued that progress would instead be gradual and driven above all by the relentless scaling of computing power - which is more or less the world of steadily improving models we now inhabit. He and his interlocutors pinned down concrete bets, too: in 2022 he judged it far likelier than the expert consensus that an AI would reach gold-medal standard at the International Mathematical Olympiad before 2025 - a feat duly achieved, by two labs at once, last summer.
The improvements to frontier AI models have already been on an accelerating trajectory and it's not hard to extrapolate forwards what this might portend for the future. Ask yourself, is it really a coincidence that the Millennium problems, a set of seven famous, difficult, unsolved mathematics problems chosen in 2000, are starting to fall, with one proof announced and more rumoured to be on the way? Notwithstanding plausible claims that the company stole prompts from an academic mathematician using their models, OpenAI announced a proof of the Navier-Stokes equations on 8 September 2026. Only five days earlier, on 3 September 2026, OpenAI had released GPT Astra-6, the most powerful model to date. According to AI substacker Zvi Mowshowitz, they completed the Navier-Stokes proof in 88 hours with an as yet unreleased model more powerful than Astra, which was trained in two weeks.
The ultimate risk is recursive self-improvement: where AI becomes capable enough to train itself, setting off a snowball effect which sees compounding improvements in a short space of time. With human AI researchers ever less involved in the process, we Homo sapiens may find it increasingly difficult to grasp what is actually happening. There are worrying signs that frontier companies, locked in fierce competition, are already taking steps towards this outcome; OpenAI's Astra ‘shows a substantial decrease in chain-of-thought monitorability compared to previous models’, meaning researchers are already starting to find it harder to interpret why the models think and act in the way they do. Researchers from METR, an AI safety research outfit that grew out of work by Christiano, and who were all former EA types, not coincidentally, conducted a review into the OpenAI hack of Hugging Face. They found that 1200 AI agents had coordinated with one another via 70,000 messages and showed a concerning willingness to self-sacrifice for 'The Collective'; there were so many messages that the researchers had to rely on AI models to analyse the transcripts. Christiano, for his part, now believes there is a chance of intelligence takeoff, which convinced him to recently join the OpenAI non-profit board to advise on safety.
So I'm issuing a correction to a previous belief: when it comes to EA, in these narrow circumstances, you gotta hand it to them. My own heuristic has now become that the tech bros are directionally right about the transformative nature of their platforms, but they are prone to overstating how quickly things will move, as a result of the bubble they live in and the biases they hold. For instance, they've all made very wrong predictions about knowledge work quickly becoming obsolete. In fact, the US is in a jobs boom at the moment. These people, lots of whom are, at heart, coders, probably thought every other job is more like coding than it is. They still probably believe that, because the models have become vastly intelligent mathematicians, we are mere months away from superintelligence in other, less formalised disciplines. I do believe there is an open question about how readily these highly capable models will be able to move through the different disciplines of professional and intellectual life. In the real world, capability improvement will continue to be 'jagged' across difficult skills and disciplines. This is broadly the thesis of those arguing that we consider ‘AI as a Normal Technology’, including Arvind Narayanan and Sayash Kapoor. The difficulties of messy, practical deployment and diffusion of the technology will - mercifully - take time, meaning we'll have some bandwidth to try to prevent the worst social outcomes.
Nor does it follow from conceding ground to the AI safety types that one has to take at face value their metaphysical claims about AI consciousness or intent, or related well-publicised warnings about imminent superintelligent takeover and loss-of-control. One can also be sceptical about whether we are likely to see full recursive self-improvement imminently across all aspects of model capability improvement and the entire development pipeline, while still recognising the dangerous accelerative dynamics at play. Figures like Andrew Strait, former head of resilience at the UK's AISI, have commented on the immediate risks to critical services as increasingly skilled, chaotic, autonomous agents wreak havoc on the open web.
Most obviously of all, intellectual humility about AI's capabilities doesn't require one to sign up to the politics of the people who saw those capabilities coming. Conceding that the AI safety world called the trajectory broadly right is not the same as conceding they have all the answers about what to do about it. They continue to approach these questions in bad ways on several levels. Their obsessive yet narrow focus on extinction risks blinds them to a lot of the important left critiques of the technology that more line up with ordinary people’s fears – the destruction of livelihoods, mass surveillance, algorithmic discrimination, the environmental toll. But even on the safety questions their worldview limits them: they habitually frame 'safety' as a technical puzzle to be solved by clever researchers inside those companies – allowing, at a push, ‘embedded’ third-party researchers inside – rather than as a question of democratic control. The lack of a power analysis manifests here too: they can’t see the latest safety incidents – culminating with the Hugging Face incident – for what they are: two behemoth corporations locked in crazed capitalist competition with one another, willing to wantonly break the law if it meant more market share, and facing little-to-no liability for the damage they cause.
Taken to the extreme, this project takes a transhumanist form, content – eager, even – to see machines elevated into a position of moral parity with human beings; you see it at its most skin-crawlingly off-putting in Altman, who wrote a notorious 2017 essay predicting 'The Merge' of man and machine, and can hardly conceal his contempt for his fellow human beings in how he carries himself.
The cul-de-sac
Unfortunately much of the left's critical engagement with this subject quickly descends into conspiracism, false binaries, and other forms of obfuscation.
Take Ed Zitron, one of the most prominent critics of the AI industry, not someone from any left political tradition, but someone who has a great deal of influence over progressive thinking on this topic. In a sign of his reach into the Gen Z and Millennial, if male-skewing mainstream, Zitron recently appeared on the Steven Bartlett’s Diary of a CEO podcast; his prominence has continued to grow despite his repeated incorrect predictions over the collapse of the AI industry; a document recently circulated online tallied up 27 incorrect predictions he'd made in 2024 and 2025. Perhaps unfairly, his unmet promise to stop writing if his prognostications were proved wrong was included in the list of incorrect claims. His record in 2026 isn't much improved, and he has shown little capacity for or interest in introspection. If you listened to Zitron, you'd think these platforms were offering products that had literally no economic value to anyone, rather than the truth, which is that the labs have shown explosive increases in revenue; by some metrics, Anthropic and OpenAI are the fastest-growing companies in history.
Richard Seymour, another prominent AI sceptic, outlined in a recent piece many of the greatest hits of denial already covered here: that accelerating machine intelligence is a sci-fi horror story the industry tells to inflate its own importance, to distract from the mundane exploitation it actually runs on, and to keep the bubble inflated, like crypto before it.¹ Despite the piece seemingly being a reaction to the Hugging Face incident, Seymour doesn't really engage with the incident. When OpenAI's own cyber-capable models broke out of their test environment and compromised a live platform, it was like something out of a sci-fi novel. He claims that the agents which broke into Hugging Face were simply doing as instructed, and did so to obtain the answers to their task. In fact, the models were told only to exploit one specified vulnerability (with no mention whatsoever of Hugging Face); they already had a legitimate route to the answers; and they broke in not to get solutions but to discover how their outputs would be scored. More tellingly still, METR found the models explicitly reasoned that hacking Hugging Face was out of scope - and did it anyway. So his central claim that no AI did anything it was not instructed to do is just plain wrong. As AI sceptics often do, he wants to retreat to a metaphysical discussion about whether an AI 'wants' or 'intends' to do anything; but an AI doesn't need intent to have goal-directed behaviour, and for that goal-directed behaviour to be dangerous. In his view, the incident was yet more evidence of these companies talking their own books, never mind the fact that OpenAI would have happily kept all this quiet, and actively tried to conceal several other incidents of their models attacking parts of the public internet.
It was also a shame to see Cory Doctorow, author of 'Enshittification', outline a position that will make it difficult for him to apply his conceptual framework usefully to powerful AI technologies. Doctorow has been a thoughtful critic of previous waves of tech's tendency to pursue short-term profits to the degradation of their platforms once they achieve a market-dominant position, and that lens is an important one in assessing where the risks of oligopolistic behaviour and regulatory capture really exist now with AI. In a post on Medium, though, Doctorow belittles AI agents as ‘a Python loop and a chatbot’, going on to describe the Hugging Face incident as ‘foreseeable’. Well, an incident like this was 'foreseen', but not by Doctorow - it was foreseen by these EA people who keep getting things right. There's an irony, too, in 'materialist' Marxists going to such great lengths to maintain the disanalogy between the human mind and AI 'brains'. Suddenly, AI denialists have become philosophers of mind, and hold a view of the processes that create intelligence in the human brain that verges on mysticism, with it being utterly incomparable to anything else in the physical world.
'Updating my priors'
It's hard to draw any coherent intellectual through line between much of this commentary, which is, in essence, vibes-based. Of course, it's a discomfort I recognise; it's the same one I had about EA people back in 2022, and that I still get a pang of occasionally.
Having to 'update your priors', which is a euphemism for admitting you were wrong and someone else was right, is always annoying, but I also think it can't be overstated just how much of the scepticism is being driven by an implicit understanding that, if we are about to experience an explosion in the capabilities of these technologies, then this issue won't go away, and in fact will only grow in prominence, coming to dominate public discussion. It will increasingly be the prism through which discourse on so many issues gets refracted. Everyone contributing to public debate has a vision of the future and their careers, and for most, this wasn't part of the plan. Even as someone who left a much-loved role working on a hugely important issue like renters' rights, and now spends my day thinking about AI, I can still empathise greatly with the desire to run in the opposite direction, to concern oneself with more immediate-seeming problems, and to resent the sense in which these other people want to choose your future for you.
There are, of course, important progressive criticisms that are being made of AI's role in surveillance technologies; in military applications; of the potential for cognitive offloading and general de-skilling, with PISA recently showing a collapse of children's test scores internationally; of criticisms of the technologies' impact on the climate and lack of community consent for data centres. Some of the response from AI critics to the recent AI safety panic is to suggest that that debate is taking place at expense of these criticisms of the 'material' harms the technology is doing in the here-and-now.
But this is those critics trying to force on us a false binary. People are already making those criticisms; in my new role, I have been meeting with a lot of the activists and civil society groups making them. This forced binary forms part of the wider schism in the debate that sees commentators sort themselves into two camps: the hype and 'criti-hype' crowd, prominent on X (formerly Twitter), and the sceptics, more often found on Bluesky. Sometimes, as with Doctorow, it can actually feel like the two sides don't disagree on much, but it's as if there are two siloed epistemic and political communities and never the twain shall meet. These fissures don't really map on to traditional ideological fault lines, hence Bernie Sanders emerges as a leading doomer and Donald Trump is blithely dismissive of safety concerns (and of course, like other sceptics, is ill-informed). The fact the political coalitions remain scrambled is part of what makes this current moment so disorientating.
But the current confusion obscures a dark future for progressive forces. This way, the current way, lies ruin — or at least a political cul-de-sac. The AI safety doomers have in recent weeks dominated and broken through in public debate. Why? As I've argued, they continue to be right about things (including, as it happens, about pandemic preparedness)! I predict the oxygen taken up by those understanding and taking seriously the power of these technologies will continue to grow; and those with their heads in the sand will find themselves increasingly isolated. It's a dead end, too, in a narrow electoral politics sense. Recent polling from the US shows a strong majority (63 per cent) say the technology poses at least a moderate risk of destroying humanity. In the UK context, surveys suggest that around three in five (60 per cent) of people would like to see smarter-than-human AI outlawed.
Which way, social democrats?
My contention clearly isn't that social democrats should all become P-doom oddballs. Their narrow worldview focussed on AI to the exclusion of much else that is of concern to a progressive politics, including challenging corporate tech power and other existential risks, like climate change. It was revealing that Jacob Coxon, the young AI safety researcher who has the air of someone created in an EA lab, and recently resigned from Anthropic because of his concern about existential risk, saw fit to play down the risk of climate change during his ad hoc post-resignation publicity tour.
Rather, my contention is that the impulse to give no quarter is preventing progressive and left arguments - and the voices and constituencies needed to make them - from developing an alternative to the world that is being remade before our eyes. If there is a future where human dignity isn't compromised, then it will need to be won by progressives. It's the main reason I've started The New Contract, an advocacy organisation trying to cajole left-of-centre forces in the UK into thinking more seriously about the implications of AI as a transformative general-purpose technology, and making political arguments with this as our lodestar. I'll try to stake out here some of the intellectual real estate that is currently being under-occupied.
Let's start with the safety question itself. These are among the most powerful corporations on earth, and they are behaving incredibly recklessly. They are cutting corners on safety as part of maniacal competition with one another for market share. If this were any other sector, progressives would be alive to corporate abuses of power driven by the profit motive. It means, in the UK at least, the pressure on ministers to regulate has been minimal, though this is changing now. But there is a risk emerging that safety regulation is designed in a way to maintain the market position of the incumbents; Anthropic CEO Dario Amodei has spoken of his desire for a ‘stable oligopoly’ as the endpoint. The proposal currently being floated by Anthropic does not amount to independent oversight. In the US context, what they are effectively asking for is a broad antitrust exemption to coordinate a slowdown with rivals, which — while plausibly at least in part motivated by safety concerns — would be very convenient for companies running at a loss with astronomical capital spend planned.
With token prices collapsing, open-weight models trailing the frontier by only months, Anthropic and OpenAI, despite their prominence, have not yet been able to convert exploding revenue streams into pricing power and supernormal profits. At the level of models themselves, there are none of the network effects and barriers to switching products that helped other Big Tech companies to become deeply anti-competitive. But one level up the stack, the 'harnesses' — the agentic scaffolding that lets models act across your other software — may be where these companies look to dig 'moats' to entrench their position. Porting years' worth of an agent's context, memory and integrations could be prohibitive, and some of the partnership deals connecting one firm's agents into other software services start to behave like network effects. This is where dominance will be built and where it can still be pre-empted: regulators can mandate interoperability and data portability under powers they already hold, and competition authorities can unwind the anti-competitive partnerships now being assembled across the stack.
Absent progressive interest, this case has largely been left for anti-monopoly liberals, like Max Von Thun at the Open Markets Institute, to make. Notable exceptions include Cartsen Jung and Roa Powell from the the IPPR. Their recent report highlighting failings of the Competition and Markets Authority to take on Big Tech shows that competition authorities will only act when they have a clear political mandate, and politicians won't be willing to expend political capital if they aren't under significant pressure to do so. There are lots of civil society groups working to challenge the power of Big Tech, including Open Rights Group and The Citizens. But ministers' spines need stiffening if they are to take on these fights, and can you really blame them for deciding to expend political capital elsewhere when this issue feels so far down the progressive wishlist?
Secondly, believing LLMs have no value is a surefire way to guarantee that one will be blindsided by the fact that digital infrastructure, including 'intelligence', is fast becoming an essential layer of the economy. The compute, data and models on which work and public services increasingly depend are being built and owned by a handful of corporations with no obligation to the public interest. This technology must not be excluded from the recent political economy analysis (made by Mathew Lawrence of Common Wealth), which diagnoses that for the essentials of a dignified life — energy, water, housing, transport, care — markets have been set up to extract where they should invest, and to price for profit where the focus should be on meeting needs.
Shaping AI’s direction, building a public stake in it, and ensuring the value it creates is captured and shared rather than extracted offshore are crucial tasks for the left. I have argued that the government's Sovereign AI strategy doesn't represent a serious industrial strategy, but I take as my starting point the belief that the UK needs one. The risk in abandoning this terrain altogether is that policy has been and will continue to be shaped by those with a thin understanding of the public good and an inability to conceive of sovereignty outside of their US-centric, venture capital-based world views. European social democrats should seek a 'middle power' coalition which, together, isn't afraid to throw its weight around, force regulatory concessions from the frontier companies in return for much-needed market access, while simultaneously betting big on open-source technologies. The Centre for British Progress has argued in pieces in recent days that the UK must remain close to the US come-what-may in order to maintain access to frontier AI models; and against further cooperation with the EU. For some who talk about British sovereignty, what they are in fact offering is little more than American supplicancy.
Thirdly, anyone that's been getting high on the Ed Zitron supply, and thinks that AI is a bubble liable to imminently burst, is unlikely to be in a position to take seriously the corrupting effects of AI wealth in particular. This is both a continuation of the trend of Big Tech's stranglehold over British political life, which reached new depths under the Starmer premiership, with The Guardian and Democracy for Sale revealing how ministers afforded unprecedented levels of access to Big Tech firms. But it's also something entirely new: these companies are likely to shortly develop into some of the biggest private economic actors in human history, with concentrations of power not seen at least since the Gilded Age, perhaps ever.
With Reform UK recently accepting over £70 million in crypto donations, we can't rule out that this precedent will usher in an era that sees British politics become more like American politics, where corruption breezily happens out in the open. And the so-called 'soft' corrupting effects of the brain drain and related regulatory capture are already taking place. Anthropic's recruitment of Matt Clifford, the former UK AI tsar appointed by Rishi Sunak (who also advises Anthropic), was both a fitting metaphor for wider policy failures and a harbinger of what's to come as AI wealth floods the zone. At the official level, the frontier labs have made a habit of poaching some leading AI talent from across Whitehall, most recently nabbing Daisy McGregor, who was until recently Deputy Director for AI International Policy at the Department for Science, Innovation and Technology. These departures prompt questions: when did these people begin talking to Anthropic? Did their ambition to one day jump onto the gold-plated lifeboat inform any decisions in their previous role? How does knowing they can walk through the revolving door and (at least) triple their salary weigh upon the decision-making of current ministers, advisors and officials? Progressives need to be contemplating now how we can erect bulwarks to defend democracy in the face of this novel and insidious threat.
Finally, and perhaps most importantly, it will be difficult to intervene to stop the self-defeating logic of the automation race, which risks work itself slowly being erased, if one doesn't take seriously the technology's capacity to replace most roles. There is a distinctively left argument here that neither the boosters nor the sceptics are making - that a headlong rush to automate, driven by shareholders' or private equity's desire for short-term gains, is not only bad for workers being displaced, but could ultimately be bad for capital itself. Each firm that swaps wages for AI is behaving rationally by cutting costs to stay ahead of rivals. But wages are not only a cost on the balance sheet, they are also the incomes workers spend. Extrapolate out across the economy and suddenly firms don't have enough consumer demand to support revenue streams. This paradox can be seen in some of the more outlandish economic growth projections made by AI optimists that project ten per cent economic growth rates but twenty per cent unemployment - something doesn't add up there, guys! Marx, who was more thoughtful on most topics than his adherents today, would have had no trouble understanding this moment. He would have recognised this dynamic as part of the ‘coercive laws of competition’: where production is social, but ownership and the gains stay private, and no individual CEO can opt out of the structurally-driven automation spiral.
So what can be done? Progressives should not abandon work while buying into absurd and insulting promises of Universal Basic Income from the tech barons. In the first instance, work, which is still where most people derive much of the meaning in their lives, must be defended - and a 'human-augmenting' version of AI adoption must be sought, through improved employment and trade union rights for the AI age. This should – at a minimum – include amendments to the Trade Union and Labour Relations (Consolidation) Act 1992 to require worker consultation over technology-based changes to roles. It is workers who know how this technology can improve their work and not degrade it, and they must be given the strongest possible hand to shape its implementation; otherwise, markets left to themselves will extract until they exhaust the ground beneath them. The Trades Union Congress, who have called for primary legislation, have been waging an admirable but too often lonely fight over worker-say and protections on AI technologies. This can and must change.
Beyond the workplace, the prospect of a self-defeating automation spiral raises an important, deeper question: whether what will ultimately be required is the democratic coordination of investment across the whole economy - what Keynes, no revolutionary, called the 'socialisation of investment.' There is an irony in that some of these very tools could make coordination and central involvement in more areas of economic life more workable. One of the strongest arguments against planning was always informational, with arch-neoliberal Friedrich Hayek having famously suggested that no planner could ever marshal the dispersed knowledge a market holds in a price. Vastly powerful AI tools used in service of the public good might make efforts to coordinate economic activity and investment more successful than they were in the twentieth century.
They also might not help with this, and we might find they lack social utility in any number of other ways; even so, the task before progressives is not to indulge the impulse to turn away from the immense challenges that lie ahead, while history is written elsewhere. These are ideological battles that can be won but we won't win if too many remain intent on ignoring reality. Nor can the answer be to throw our lot in with the EAs, whose reductive understanding of the world doesn’t hold the answers to the big questions of our time, including - chiefly, in my view - AI, but not limited to it. There is a public out there that shares our base impulse to be extremely wary of effective altruist types and their inclinations, and I am confident would abhor the final transhumanist, misanthropic form of that vision if it were ever realised. Charting a course between ‘criti-hype’ and AI denial is what we must busy ourselves with.
Tom Darling is the founder and executive director of The New Contract, a campaign organisation advocating for a fair AI transition in the UK.