The UK's Regulating for Growth Bill: AI through sandboxes, not a statute

When the British government finally put an AI bill into its legislative programme, it was not the frontier-safety statute that had been trailed for two years. In the King's Speech of 13 May 2026 the headline vehicle for AI was the Regulating for Growth Bill — a cross-economy, pro-innovation bill whose stated purpose is to "reduce the burden of unnecessary regulation through innovation". For anyone reading the UK against the EU AI Act, the direction of travel is the point: where the EU regulates AI ex ante through conformity assessment, the UK is legislating to make it easier to deploy.

What the bill does

The background briefing notes describe a bill that "builds on the 2025 Regulation Action Plan to modernise regulation so it supports growth and innovation while maintaining essential safeguards". Three mechanisms carry that aim.

First, sandbox powers. As announced by the Chancellor in her 2026 Mais Lecture, the bill creates cross-economy "sandboxing powers" — legal powers to temporarily relax existing rules, under strict controls, so businesses can test new products and technologies in real-world settings. If a trial proves successful, the bill allows the change to "quickly be embedded permanently into law". The notes single out AI explicitly: the government will be "exploring cross-cutting AI sandboxes, enabling responsible testing and adoption of AI-enabled products and services across multiple sectors where existing regulatory frameworks currently slow innovation".

Second, a strengthened growth duty. The bill gives a list of leading regulators — Natural England, the Environment Agency, the Health and Safety Executive among them — a statutory mandate to prioritise growth without undermining their core functions, backed by reporting requirements so the duty has measurable effect.

Third, strategic steers. Ministers gain a new statutory power to issue strategic steers, defining what "growth" means in different regulatory contexts — a lever to direct regulators, including the ICO, FCA, CMA, Ofcom and MHRA that already carry the UK's AI principles, towards enabling innovation.

The government frames trials as operating under "strict safeguards, including protections for consumers, workers and human rights, with clear accountability and regulatory oversight throughout", and insists the legislation "is not about deregulation". It points to the FCA's own 2016 'Innovate' sandbox, Singapore's Pro-Enterprise Sandbox and Canada's experimentation frameworks as the model.

Why it matters for AI governance

The significance is what is absent. The much-discussed government frontier bill — mandatory pre-deployment evaluations, incident reporting, binding obligations on the largest models — did not appear in the King's Speech. The legislative slot went instead to a growth-and-sandbox vehicle, and the AI Security Institute remains a technical body rather than a regulator with statutory powers. The UK has, for now, decided that its statutory AI intervention should lower regulatory friction rather than raise it.

That leaves the British model resting on three legs in 2026: the non-statutory five principles applied by sector regulators (see the UK approach to AI); the one genuinely hard, cross-sector obligation, which arrived through data-protection law (see the UK's statutory AI and ADM code); and now a bill whose AI content is a permission to experiment, not a constraint. Set against the EU — where the AI Act timeline and the Digital Omnibus debate turn on how much ex-ante duty to impose — the contrast is stark. For an international organisation the AI Act remains the governing upper bound; the UK is the jurisdiction betting that agility, not conformity, is the competitive edge.