⚡ Quick Summary
Published by the Centre for the Governance of AI, this preprint chapter traces how the UK has approached AI regulation since the 2010s and argues for what should come next. It describes a distinctive UK path that is less cautious than the EU and more willing to address risks than the US, centred on a “pro-innovation” strategy of regulating AI at the point of use through existing regulators rather than regulating the technology itself.
The chapter follows the shift from the House of Lords’ 2018 report “AI in the UK: ready, willing and able?” to the March 2023 white paper, the creation of the world’s first AI Safety Institute, and the Bletchley and Seoul summits, including the voluntary Frontier AI Safety Commitments. It notes that Labour’s 2024 manifesto promised “binding regulation on the handful of companies developing the most powerful AI models”.
It then examines eight regulatory areas — barriers to adoption, frontier AI, misinformation and AI-generated content, copyright, discrimination and bias, biological design tools, AI agents, and AI-driven unemployment — and proposes a flexible, principles-based regulator for frontier developers, defined initially by training-compute thresholds and imposing safety, cybersecurity and transparency obligations.
🧩 What’s Covered
The chapter moves from history to prescription, in the document’s own order.
- Introduction and framing: poses the question of what “needs to be done” when understanding of AI’s future is limited, quoting Turing’s 1950 remark and Peter Kyle’s January 2025 framing of the AI Opportunities Action Plan.
- History of UK AI regulation: from Babbage and Lovelace through the Colossus computers, the 2012 AlexNet result and DeepMind’s founding, the 2016 Commons science committee report, the 2018 Lords report, the 2021 National AI Strategy and the 2022 “pro-innovation” policy paper.
- 2022–2024: catastrophic risk and a hands-off approach: ChatGPT’s arrival, the March 2023 white paper, the £100 million AI Foundation Model Taskforce, the Future of Life Institute letter calling for a training pause, and the Bletchley Declaration endorsed by 29 signatories.
- The AI Safety Institute and international summits: AISI’s mandate to run technical evaluations, the Seoul Declaration and voluntary Frontier AI Safety Commitments, and the first International Scientific Report on the Safety of Advanced AI.
- Labour’s objectives: the five missions, the highest sustained growth target for the G7, the AI Opportunities Action Plan’s roughly 50 recommendations, and the Regulatory Innovation Office.
- Frontier AI regulation: the case for regulating development, compute thresholds around 10^26 floating point operations, and three sets of obligations covering safety, cybersecurity and transparency.
- Content, copyright and discrimination: deepfake criminalisation, watermarking and C2PA content provenance, the copyright consultation and opt-out text-and-data-mining proposal, and updating anti-discrimination frameworks for algorithmic decisions.
- Biological design tools, AI agents and unemployment: dual-use bio risks and DNA synthesis controls, agent accountability loopholes and fiduciary duties, and labour-market monitoring, AI Growth Bonds and scenario planning.
💡 Why it matters?
UK AI regulation remains unsettled: the government has committed in principle to legislating on the most powerful models but, at the time of writing, has not done so. This chapter gives policymakers, compliance leads and safety teams a compact account of the institutional history behind that commitment and a concrete menu of design choices — regulator mandate, compute-based scope, obligations, and alternatives to point-of-use regulation. It also connects UK choices to the EU AI Act’s general-purpose model rules, the G7 Hiroshima process and the Frontier AI Safety Commitments, which is useful for anyone assessing whether a UK regime would be interoperable with obligations elsewhere.
❓ What’s Missing
The chapter is a policy argument rather than a compliance text: it does not map UK duties onto specific statutory provisions, offer templates, or quantify the costs of the proposals it advances. Several questions are left open by the authors themselves, including whether the AI Safety Institute should become the frontier regulator, how compute thresholds should be adapted over time, and how reliable watermarking can become. Coverage stops in early 2025, so subsequent legislative steps are not addressed, and the analysis is bounded to the UK legal order apart from comparative passages on the EU and US.
👥 Best For
Best for policy and regulatory affairs teams tracking UK AI legislation, compliance leads comparing the UK’s sectoral approach with the EU AI Act, and safety or evaluation staff who need the rationale behind compute thresholds and frontier safety commitments. Also useful for researchers and journalists wanting a sourced timeline of UK AI policy from 2010 to early 2025.
📄 Source Details
From Turing to Tomorrow: The UK’s Approach to AI Regulation by Oliver Ritchie, Markus Anderljung and Tom Rachman, Centre for the Governance of AI. The first page is an arXiv preprint dated 3 July 2025 (arXiv:2507.03050v1) and states that it is a chapter prior to acceptance for publication in a forthcoming book, without editorial or peer review. 32 pages, English. The extracted text covered all 32 pages. No URL for the document itself is printed; the reference list cites other works.
