The UK Is Writing the Rules for AI in Healthcare Right Now, Not Later

A regulatory affairs lead at a UK pharma company gets a request: can the medical writing team start using an AI tool for drafting clinical documents? The honest answer, right now, is that nobody can point to a finished UK framework and say “here’s the rule.” The rulebook is still being written, in public, with input still being gathered. That’s not a gap in someone’s knowledge. It’s the actual state of UK AI regulation in healthcare in 2026.

What We Actually Know

The Medicines and Healthcare products Regulatory Agency has been visibly active on this front. It launched a call for evidence on AI regulation in UK healthcare, seeking input from clinicians, industry, and the public. That led to the creation of a National Commission into the Regulation of AI in Healthcare, bringing together global AI leaders, clinicians, and regulators specifically to advise the MHRA on building a new regulatory framework.

Alongside that, the MHRA has been running an AI Airlock, a regulatory sandbox built with the NHS, designed to pilot AI models and software as a medical device in a controlled setting before wider rules exist. New post-market surveillance legislation for medical devices took effect in June 2025, strengthening data-gathering obligations that will likely extend to AI-integrated products.

None of this is speculative. It’s active, funded, and moving. The MHRA secured £7.5 million in new research grant funding this year specifically to advance regulatory science and support innovation across medicines and medical devices, AI included.

What’s Genuinely Still Unsettled

Here’s the part worth being honest about: there is no finished UK framework for AI in healthcare yet. The Commission exists to advise on building one, which means, by definition, it doesn’t exist in final form today. Standards for AI as a Medical Device are still under development. The Airlock is a pilot, not a permanent structure.

For a pharma or life sciences organization asking “are we compliant,” the honest answer is that compliance is being defined while your team is already using these tools. That’s an uncomfortable position, but pretending otherwise, either by assuming the UK is unregulated territory or by assuming a clear rulebook already exists, are both wrong in ways that create real risk.

What To Do in the Meantime

Treat UK AI use in regulated workflows as if scrutiny is coming, because it is. The direction of travel is unmistakable, more oversight, more formal standards, more post-market data obligations, even if the exact shape isn’t final.

Choose AI tools now based on the data handling standards likely to matter later: clear data residency, no training on submitted content, auditable usage, rather than waiting for a finished rule to force the decision retroactively.

Watch the Commission’s output directly rather than secondhand summaries, since this is a live process and secondary commentary can lag or oversimplify what’s actually being proposed.

Don’t assume UK divergence from the EU means lighter treatment. The MHRA’s activity level here suggests the opposite, a UK-specific framework built from scratch, informed by global experts, rather than a lighter copy of EU rules.

Conclusion

The UK isn’t behind on AI regulation in healthcare, it’s actively building it, in public, right now. For pharma and life sciences organizations operating there, the smart move isn’t waiting for a finished rulebook. It’s choosing tools and practices today that will still hold up once one exists.

Key Takeaways

• The MHRA has launched a National Commission specifically to advise on a new UK AI regulatory framework for healthcare, this work is active, not hypothetical.
• No finished framework exists yet, standards for AI as a Medical Device are still under development.
• The MHRA’s AI Airlock, run with the NHS, is piloting real AI oversight ahead of formal rules.
• UK divergence from the EU AI Act doesn’t mean lighter treatment, the MHRA’s activity suggests a rigorous UK-specific framework is coming.
• Organizations should choose AI tools now based on the standards likely to matter later, not wait for final rules to force the decision.