AI and patient safety: The risks, the opportunities and what needs to change

Summary

AI is already reshaping healthcare, from ambient voice scribes in GP practices to patients self-diagnosing through large language models before they reach a clinician. But the question of whether it is making care safer or introducing new risk is not being answered clearly enough. In this session from the Connected Health & Care Summit 2026, Mark Linggood of RLDatix and Clive Flashman, Chief Digital Officer at Patient Safety Learning, hold a frank discussion on the evidence, the regulatory gaps and the role AI could play in improving incident reporting itself. 

Watch a candid discussion between RLDatix and Patient Safety Learning on how AI is affecting patient safety, what the research shows, where the regulatory gaps exist and how AI could transform incident reporting. 

Content

This fireside chat covers both the risks and opportunities of AI in healthcare, grounded in recent research and practical experience. The key points discussed were: 

• A Canadian Medical Association study found patients who followed AI advice were five times more likely to experience harm, and 97% of physicians had personally intervened to prevent harm from AI or online advice.

• An Oxford University study found that the models themselves were generally accurate, but the human-model interaction was the primary source of risk, with a difference in one word in a symptom description producing life-or-death differences in advice. 

• AI models are now capable of hallucinating citations, generating fake references that appear to come from journals like the BMJ but do not correspond with genuine, published sources.

• The UK’s current reporting systems and regulatory frameworks are not designed for an AI-enabled world, with LFPSE categorising AI-related incidents only as generic “IT issues”.

• The Oxford study reported that none of the tested models is ready for deployment in direct patient care.