The coming AI crisis: UK preparedness and response
The government's own analysis says about 70% of British workers are in jobs exposed to AI, a higher share than America's. Its policy answer is to accelerate: more compute, more adoption, no new law. Graduate job postings have already fallen sharply, and no AI bill is before Parliament. Britain is preparing thoroughly for AI's upside and barely at all for its downside.
The gaps below are the missing shock absorbers: nobody watching displacement as it happens, no income bridge for workers whose jobs contract, no incident register, no election rules for deepfakes. Most are cheap by Whitehall standards. All get harder to build once the crisis is visible.
Full landscape notes (July 2026)
Mid-2026 UK AI policy is dominated by the growth agenda. The AI Opportunities Action Plan one-year-on review (Jan 2026) reports 38 of 50 commitments met: public compute up from 2 to 21 ExaFLOPs, five AI Growth Zones designated, a £500m Sovereign AI Unit entering its next phase, and the 'Humphrey' toolset spreading through Whitehall. But the build-out rests on US capital and control: the September 2025 Tech Prosperity Deal brought ~£31bn from Microsoft, NVIDIA, OpenAI and CoreWeave, and the CMA found AWS and Microsoft each hold 30-40% of UK cloud. The AI Security Institute is world-class (100+ technical staff, 30+ frontier models tested) yet entirely non-statutory, and the promised Frontier AI Bill has slipped repeatedly; no AI bill is before Parliament. Meanwhile labour signals have turned: youth unemployment around 16%, graduate postings reported down ~45% year-on-year, call-centre employment down 19% (2021-25), and King's College London finds AI-exposed firms cutting junior roles by 5.8%. DSIT's own January 2026 assessment says roughly 70% of UK workers are in AI-exposed roles (higher than the US), with a third of the workforce in the high-exposure/low-complementarity (displacement-risk) segment. Transition policy is thin: skills programmes exist (Youth Guarantee, TechFirst), but there is no displacement observatory, no income-bridging instrument, and the only UK income-floor evidence (the Welsh care-leavers pilot) reports finally in 2027. Election law, liability and incident reporting all lag deployment. The posture is: accelerate adoption, partner with US labs, defer regulation. That leaves downside preparedness largely unbuilt.
The gaps (11)
No real-time observatory of AI-driven labour displacement
AI is reshaping British jobs right now. The state finds out years later.
No transition instrument for AI-displaced workers: the UK has no TAA equivalent
Call-centre jobs are down 19% and there's no plan for the people losing them.
AISI has no statutory powers while the Frontier AI Bill slips indefinitely
Britain tests the most powerful AI models only when their makers say yes.
No central AI incident reporting and investigation function
Plane crashes get investigated. Drug reactions get logged. AI failures just vanish.
Election law does not cover AI-generated content and no regulator owns political deepfakes
Political deepfakes are legal, unlabelled and already reaching three in ten adults.
Public-sector AI deployment is outrunning assurance, transparency and evaluation capacity
Government is Britain's biggest AI adopter and its least audited.
No fiscal early-warning capability for AI-driven tax-base erosion
No one in government models what AI displacement does to the tax take.
No place-based transition plans for AI-exposed local economies (contact-centre towns)
The towns hosting AI data centres could lose their biggest service employers to AI.
Liability for autonomous AI harms is unresolved and no reform vehicle exists
An AI agent can cause harm that no person is legally liable for. That's the law now.
Entry-level pipeline collapse: no instrument keeps firms training juniors
Graduate job adverts are down nearly half. The real bill arrives in the 2030s.
No decentralised attestation network for critical public facts
When anything can be faked, Britain still proves 'what happened' with a single web page.
Who is already here: key actors (15)
- AI Security Institute (AISI) (government body): DSIT directorate with £240m funding to 2030 and 100+ technical staff; has tested 30+ frontier models under voluntary access agreements; no statutory powers, and its remit was narrowed to security risks (dropping bias/societal impacts) in the February 2025 rename.
- Sovereign AI Unit (DSIT) (government body / funder): Up to £500m to build sovereign AI capability (compute allocation, datasets like OpenBind); next phase launched April 2026, chaired by venture capitalist James Wise.
- AI and Future of Work Unit (DSIT) (government body): New cross-government evidence unit on AI labour-market impacts; first assessment (Jan 2026, with AISI) found ~70% of UK workers in AI-exposed roles and admitted the evidence 'does not yet provide clear answers' for policy.
- Incubator for AI (i.AI) / Government Digital Service (government body): Builds and deploys the 'Humphrey' suite (Consult, Parlex, Minute, Lex, Redbox, Extract) across the civil service: the sharp end of public-sector AI adoption.
- AI Research Resource (Isambard-AI, Bristol; Dawn, Cambridge) (infrastructure programme): The UK's public AI compute: Isambard-AI (5,448 GH200 chips, launched July 2025), Dawn expanding sixfold in 2026; £2bn committed to expand AIRR 20x by 2030 plus a £750m Edinburgh national supercomputer.
- AI Growth Zones programme (government programme): Five designated zones (Culham, North East, North Wales, South Wales, Lanarkshire) with fast-track planning and power; government claims £28.2bn private investment and 15,000+ jobs, mostly US-hyperscaler-anchored data centres.
- AI Growth Lab (DSIT) (government programme): Cross-economy regulatory sandbox launched 8 June 2026 (legal services/conveyancing first) allowing licensed, supervised relaxation of rules for AI pilots, the government's chosen substitute for an AI bill.
- Skills England (government body (DfE arm's-length)): Publishes annual skills reports and sectoral needs assessments including AI; administers levy-funded training landscape but has no AI-displacement-specific remit or instrument.
- Electoral Commission (regulator): Launched a deepfake-detection pilot ahead of the May 2026 elections; has itself called for the digital imprint regime to be extended to all online election material, a change only government can legislate.
- Ada Lovelace Institute (research institute): Nuffield Foundation-founded; the leading independent analyst of UK AI regulation, public-sector AI deployment and societal impacts, consistently documenting the gap between adoption and governance.
- Institute for the Future of Work (IFOW) (charity / research): Ran the Pissarides Review into the future of work; publishes evidence on AI's effect on entry-level and graduate jobs and advocates algorithmic impact assessments in employment.
- Centre for Long-Term Resilience (CLTR) (think tank): Authored the incident-reporting and frontier-AI-governance blueprints (central incident database, 'three lines' risk model, AISI statutory powers) most cited in this space; they remain largely unimplemented.
- Centre for Emerging Technology and Security (CETaS, Alan Turing Institute) (research centre): Key work on AI-enabled influence operations against UK elections and on the UK-US tech alliance's strategic dependencies.
- Nscale (company): UK AI-infrastructure firm behind Stargate UK (with OpenAI and NVIDIA, sited in the North East AI Growth Zone) and Microsoft's Loughton AI campus: the main domestic corporate vehicle for the US-financed build-out.
- BritLLM / UK-LLM (UCL AI Centre, with Bangor University and NVIDIA) (project): The UK's only open-weight national LLM effort, trained on Isambard-AI and focused on UK languages (English/Welsh reasoning model, Sept 2025); academically led and small-scale, not a strategic programme.
Funders active or plausible here (11)
- DSIT / HM Treasury (£2bn compute roadmap, £500m Sovereign AI Unit, £240m AISI, £187m TechFirst)
- UKRI: EPSRC and ESRC (AIRR access programmes, Responsible AI UK, BRAID, Turing Institute core funding)
- ARIA (Safeguarded AI programme, ~£59m; technical AI safety R&D)
- Nuffield Foundation (founded and funds the Ada Lovelace Institute; funded the IFOW Pissarides Review)
- Open Philanthropy (major funder of UK AI governance/safety orgs including CLTR and GovAI)
- Founders Pledge / effective-giving funds (AI risk and policy grantmaking reaching UK orgs)
- Joseph Rowntree Foundation (economic insecurity and emerging-futures work; a natural funder of transition-instrument pilots)
- Luminate and democracy-focused philanthropy (information ecosystem, Full Fact-type fact-checking infrastructure)
- Schmidt Sciences and other AI-safety philanthropies (technical safety and evaluation capacity)
- US hyperscalers as infrastructure financiers (Microsoft £22bn, NVIDIA, OpenAI/Nscale Stargate UK, CoreWeave; capital the UK leverages but does not control)
- Devolved and local: Welsh Government (basic income pilot precedent), Scottish Government, metro mayors' local growth funds (potential hosts for place-based transition pilots)
Policy notes
No AI statute exists. The Frontier AI Bill, a manifesto commitment to binding regulation of frontier developers, has slipped since 2024: first to fold in copyright, then displaced by the October 2025 regulation blueprint centred on the AI Growth Lab, a cross-economy sandbox (live 8 June 2026, legal services first) that relaxes rules rather than adding them; the Technology Secretary has signalled little appetite for new regulation beyond extending online safety. Governing law is UK GDPR, the Data (Use and Access) Act 2025 and sector regulators. The Online Safety Act omits political deepfakes and democratic-harm misinformation; the digital-imprint regime binds only registered campaigners. The CMA completed its cloud investigation (July 2025) finding AWS/Microsoft duopoly concerns, then shelved SMS designation. The Law Commission's July 2025 liability paper has no reform project behind it. AISI remains voluntary and non-statutory. Employment and fiscal frameworks are effectively AI-blind.