UK Government publishes updated AI scenarios for 2030
The future of artificial intelligence is anything but certain – and the UK government knows it. In June 2026, the Government Office for Science (GO-Science) released an updated set of AI Scenarios 2030, designed to help policymakers navigate the profound uncertainties surrounding AI's trajectory. This isn't a prediction, mind you. It's a toolkit for stress-testing policy against a range of plausible futures.
The original scenarios were developed in 2023 and published in April 2025. But the AI landscape has shifted dramatically since then. Capabilities have skyrocketed, investment has poured in, adoption has spread, and geopolitics has thrown up fresh uncertainties. So GO-Science, working with the AI Security Institute (AISI) and the Department for Science, Innovation and Technology (DSIT), went back to the drawing board. They consulted experts from government, academia, and industry to produce five new scenarios that reflect a world where AI is no longer a lab curiosity but a force reshaping daily life.
As Professor Dame Angela McLean, the Government Chief Scientific Adviser, puts it in the foreword: 'AI has advanced rapidly over the past decade. The most advanced AI systems have shifted from laboratory curiosities to those beginning to reshape our world.' The goal is simple: give civil servants a shared baseline so that long-term planning across Whitehall is consistent and coherent. Not a bad idea when you're dealing with technology that could upend entire industries.
From laboratory curiosities to reshaping the world: the rapid AI evolution
Rewind to 2023. ChatGPT had just exploded onto the scene, and generative AI was the talk of the town. Fast forward three years, and the progress has been relentless. AI systems now write code, generate images, assist in scientific research, and even help build other AI products – take Anthropic's Claude Code, which was used to create Claude Cowork, a desktop automation tool. That's recursive self-improvement in action, and it's not science fiction.
The Peterson Institute for International Economics (PIIE) blog, written by Era Dabla-Norris (IMF) and Anton Korinek (PIIE), captures the tension perfectly. They contrast the 'Washington Consensus on AI' – which sees it as a general-purpose technology like electricity, unfolding gradually – with the 'San Francisco Consensus on AI' – which expects transformative AI, possibly superintelligence, arriving on a much shorter timeline. Which one is right? The honest answer is nobody knows. And that's exactly why scenario planning matters.
The UK's GO-Science report acknowledges that plenty has changed since 2023. Investment in AI has ballooned, adoption has moved beyond early adopters, and geopolitical rivalries – especially between the US and China – have added a new layer of uncertainty. All of this made the 2023 scenarios look a bit dated. Hence the update.
The five AI futures: critical uncertainties and scenario design
The scenarios are built around six 'critical uncertainties' – the factors that will most shape AI's future but remain highly unpredictable. These are: capability (how fast will AI improve and in which domains?); distribution and model access (who controls frontier development, how concentrated will it be?); security (will AI remain controllable, and how will malicious actors exploit it?); adoption (how broadly and extensively will AI be used, how much autonomy will we grant it?); and two more factors that the full report details.
By combining these uncertainties in different ways, alongside research and expert judgment, GO-Science produced five internally consistent but contrasting futures. Each scenario is a narrative describing a plausible 2030. Importantly, they are not mutually exclusive – the real future will likely contain elements from several. And they deliberately do not assume any UK government policy intervention, because the point is to test policies against a range of possibilities.
The methodology is the same as the 2023 version, so there's continuity. But the content is fresh. One big shift: the geopolitical landscape. In 2023, the world looked different. Now, with AI being a central arena for great-power competition, the scenarios reflect that tension. For example, one scenario might see a fragmented global AI ecosystem, with rival blocs developing separate systems. Another might picture a world where open-source models dominate, making advanced AI widely accessible (for better or worse).
Divergent views on AI's economic impact: Washington versus San Francisco
This is where the PIIE analysis really dovetails with the UK government's work. The Washington Consensus says: AI is transformative, sure, but its effects will be gradual, contingent on complementary investments in skills, infrastructure, and institutions. Past innovations like the steam engine or the internet took decades to diffuse. Why would AI be different?
The San Francisco Consensus counters: because AI is different. It's not just another general-purpose technology; it's a technology that can improve itself. The speed of recursive self-improvement could compress decades of progress into years. The blog notes that 'AI systems have already become remarkably capable at programming' – and that capability feeds back into making new, better AI. If that feedback loop accelerates, the 2030 world could look radically different from today, not incrementally different.
The workshop convened by the Economics of Transformative AI (EconTAI) initiative and the IMF deliberately focused on the San Francisco Consensus scenario. Why? Because if the Washington Consensus is right, existing policy frameworks can adapt. The real danger lies in being caught off guard by rapid transformation. So they stress-tested a world where AI and robotics can perform most economically valuable tasks by 2030. That's not a prediction; it's a stress test. But it's a useful one.
Runaway adoption versus uneven diffusion: two economic scenarios for 2030
Even within a baseline of rapid AI advancement, the key variable is how fast adoption spreads. The workshop explored two scenarios. The first is uneven adoption. Big tech firms and tech-savvy hubs integrate AI quickly, but smaller businesses, governments, and developing economies lag. Regulatory friction, public skepticism, and skill shortages slow the trickle-down. Productivity improves, jobs are disrupted but not catastrophically, and fiscal and monetary frameworks remain largely intact. Think of it as a managed, if bumpy, transition.
The second is the runaway scenario. AI adoption accelerates at breakneck speed, penetrating every nook and cranny of the economy. Automation becomes ubiquitous. Jobs disappear faster than labour markets can adapt. Economic power concentrates in a handful of firms and locations. Inequality spikes. Fiscal systems creak under the strain of eroding labour tax bases and surging demands for social support. Central banks face new headaches as inflation dynamics shift. Financial markets get volatile. It's a world of winners and losers on steroids.
Neither scenario is comfortable. But the runaway one is genuinely alarming. The PIIE authors argue that policymakers need to start thinking about these futures now, not when they're upon us. Fiscal, monetary, and financial policy frameworks that worked for the past fifty years may not work in a world of rapid automation and concentrated economic power. The transition costs – social, economic, political – could be enormous.
Scenario planning as a tool for policy resilience
So what do you actually do with these scenarios? The UK government has been using them to stress-test policies across departments. The GO-Science Foresight team stands ready to support that work. For example, a scenario where AI capabilities surge but security lags might test the resilience of critical national infrastructure. A scenario where open-source models proliferate might test the effectiveness of export controls. The point isn't to pick the 'right' scenario; it's to make sure your policies are robust across a range of plausible futures.
This kind of planning is especially valuable for long-term decisions. Think about education and skills: if AI automates many current jobs, what should we be training young people for today? Or consider tax policy: if more value is created by AI systems rather than human labour, the tax base shifts. Or social safety nets: do we need universal basic income or something like it? These are hard questions, and scenario planning gives a structured way to explore them without getting paralysed by uncertainty.
Dabla-Norris and Korinek put it succinctly: 'Managing this transition will be critical to ensuring that the ongoing technical advances also advance human well-being.' They call for hedging across multiple futures rather than betting on a single one. The UK's AI Scenarios 2030 is exactly that kind of hedge – a shared baseline for cross-government thinking. It's not perfect, but it's a damn sight better than flying blind.
Closing thoughts: preparing for profound impact by 2030
Here's the thing: 2030 is only four years away. Not some distant sci-fi horizon. By 2030, the AI Scenarios report says it is 'clear that AI will have a profound impact.' Whether that impact is mostly beneficial or mostly disruptive depends on choices we make now.
The beauty of these scenarios is that they force you to confront uncomfortable possibilities. Uneven adoption could exacerbate global inequality. Runaway adoption could create social upheaval. But there are also optimistic pathways where AI augments human capabilities, boosts productivity, and helps solve big problems like climate change or disease. The scenarios don't prescribe which one we get; they illuminate the factors we can influence.
Policymakers in London, Washington, and beyond would do well to take this seriously. The Washington Consensus might be too complacent; the San Francisco Consensus might be too alarmist. The truth probably lies somewhere in between. But the cost of being unprepared for the more radical possibilities is far higher than the cost of building resilience now. As the old saying goes: hope for the best, plan for the worst. AI Scenarios 2030 gives the UK government – and anyone else paying attention – a solid framework for doing exactly that.
