MHRA opens AI sandbox to make medicines safer and faster

TL;DR:

• The MHRA has launched a regulatory “sandbox” letting companies test AI tools for predicting medicine safety alongside regulators.
• Adverse drug reactions send around 250,000 people to hospital in the UK each year, costing the NHS over £2bn.
• Up to five AI-driven approaches will be tested in the first phase, with work beginning from summer 2026.

The UK’s medicines regulator has opened a first-of-its-kind AI sandbox, announced by Science Minister Lord Vallance at London Tech Week, giving companies and researchers a controlled environment to test whether AI can make medicines safer and quicker to develop.

Targeting a costly failure rate

The Medicines and Healthcare products Regulatory Agency wants AI to improve how drug safety is assessed — better predicting risks and catching effects existing methods miss. The stakes are concrete: adverse drug reactions send around 250,000 people to hospital in the UK annually, costing the NHS over £2bn, while roughly 90% of drugs fail during development, partly because current methods cannot reliably predict how they behave in people.

Backed by the Regulatory Innovation Office, the sandbox lets innovators test tools that model how medicines are absorbed and processed, and whether they may cause harm. It will also explore using clinical data to understand effects across groups often underrepresented in studies, including children and older people. Up to five approaches will run in the first phase, with industry and academic partners from summer 2026.

The MHRA framed the work as supporting the government’s AI for Science mission, its push for alternatives to animal testing, and the 10 Year Health Plan’s goal of the “world’s most AI-enabled healthcare system”.

Looking forward

The sandbox echoes the regulator-led approach Britain is applying elsewhere, from the FCA’s financial-services AI lab to the cross-sector AI Growth Labs. For a sector where caution is non-negotiable, the test is whether a sandbox can build enough evidence to give companies confidence to invest — without diluting the safety standards that justify it.