UK opens its first AI 'growth lab', a regulatory sandbox for legal services

The government has opened applications for its first Advisory AI Growth Lab, a regulatory sandbox that lets legal-technology firms test AI tools with regulators alongside them rather than after the fact. Applications opened in early August and run until 27 September, and legal services has been chosen as the first sector to go through it. Four regulators are involved: the Solicitors Regulation Authority, the Legal Services Board, the Council for Licensed Conveyancers, and the Information Commissioner’s Office.

The point of a sandbox is to remove uncertainty. A firm building an AI tool for legal work often has to navigate several overlapping rulebooks at once, and the risk of getting it wrong is enough to slow good ideas down or stop them entirely. The Lab is meant to give joined-up, advisory guidance within the existing rules, so that innovators know where they stand before they commit. “Innovation has an important role to play in improving legal services for consumers and businesses,” said Sarah Rapson, the SRA’s chief executive, framing the exercise as support for the responsible development of AI rather than a loosening of standards. The government has pointed to conveyancing as an early use case, where AI could read a property sales pack and flag legal issues in minutes rather than hours.

It also matters as a template. This is the first of a planned series of sector-by-sector Growth Labs, and it says a lot about how the UK intends to regulate AI: not with a single sweeping statute, but by working through the practical questions one industry at a time. Legal services is a natural place to start, both because it is a large part of the economy and because it is already drawing a wave of AI startups. We wrote last week about Aavalynx, which is building AI to forecast the cost of disputes, and the directory includes Luminance, whose tools read and analyse contracts.

The reason a sandbox is worth having, rather than simply letting firms deploy and see what happens, comes down to what these tools get wrong. Many legal AI products are built on large language models, and those models can produce a hallucination, stating something false with complete confidence. In law that is not a small problem: there have already been cases elsewhere of AI inventing case citations that do not exist. A supervised environment, with the regulators in the room, is a sensible way to work out how AI can be used in legal work without putting accuracy or accountability at risk.