2025: A turning point for digital patient safety

Summary

Clive Flashman is Chief Digital Officer at Patient Safety Learning. In this long-read article, Clive shares his reflections on the digital patient safety landscape and how it has developed in the last year.

Before he looks ahead to 2026, Clive offers his insights on several critical safety topics including:

• Digital incidents, alerts and early warning
• Electronic patient records
• How not to deploy an AI tool
• Patients at the centre of digital design
• Governance, culture and skills for trusted AI

Content

From Electronic Patient Records (EPRs) and wearables to generative AI at the desktop, 2025 has felt like the year digital health stopped being ‘the future’ and became an everyday patient safety issue. Across the hub, especially in Network meetings, and on social media, the same message has kept surfacing: technology is now woven into almost every part of care. Programming is now just as important as design, governance and culture when it comes to keeping patients safe.

AI is everywhere – but it must be safe

This year has marked a clear shift from talking about ‘AI potential’ to dealing with ‘AI in use’, and with that a sharper focus on how safely AI is being deployed.

On the hub, we have highlighted the discussions from our roundtable at The Healthcare Excellence Through Technology (HETT) show discussing AI and Patient Safety. We’ve also shared how within the WHO European Region health systems are moving from pilots to scaled AI deployments, while still wrestling with readiness, regulation, and the gap between hype and reality at the bedside.[1]

Several of our 2025 hub articles have tried to unpack what ‘safe AI’ actually means in practice:

• extending Patient Safety Incident Response Framework (PSIRF) principles to digital and AI systems[2]
• strengthening AI governance
• and making accountability for harm explicit rather than implied.

In parallel, some of my articles on social media have stressed that AI is just another tool in the safety toolkit – powerful, but only as safe as the data, design and culture wrapped around it.

Digital incidents, alerts and early warning

Digital‑related safety alerts have been a regular drumbeat this year, underlining that software, data and algorithms can cause real harm if they are not designed or implemented well.

Content shared on the hub has included warnings about IT faults in maternity systems[3] and concerns about unsafe use of AI translation apps[4], both of which show how configuration and workflow decisions can have life and death consequences.

At the same time, 2025 saw the launch of a world‑first AI early warning system[5] to scan NHS data for emerging safety signals, starting with maternity. At Patient Safety Learning we have welcomed this announcement, recognising this as both a huge opportunity and a real test of public trust. Earlier detection and targeted inspections are promising, but only if data quality, transparency and clear lines of accountability are built in from day one.

Electronic patient records under the safety spotlight

By the end of the year, EPR systems were firmly in the patient safety spotlight again. We saw the publication of a new Health Services Safety Investigations Body (HSSIB) thematic report on EPR‑related safety issues[6]  and a linked response from Patient Safety Learning.

In previous years, these reports, and blogs on incidents such as the EuroKing maternity alert,[7] have shown how poor planning, rushed go‑lives and weak incident reporting can turn EPR programmes into sources of avoidable harm rather than enablers of safer care.

On social media, I have continued this conversation by commenting on botched upgrades,[8] under‑tested systems[9] and opaque vendor contracts. I have argued for robust project discipline, meaningful user testing and strong quality assurance as non‑negotiables in any EPR roll‑out.

Through roundtables and sessions with NHS England’s Frontline Digitisation team and vendors, the focus has increasingly shifted from blame to shared responsibility: national bodies, trusts, suppliers and patients all have a role in making ‘safety in use’ real.

Copilot in the NHS: how not to deploy an AI tool

Generative AI at the desktop became very real for many staff this year with the arrival of Microsoft Copilot in the NHS. In a recent blog, I described a senior patient safety manager who opened their laptop to find Copilot ‘suddenly there’. There had been limited communication, unclear governance and little practical training on how it should (or should not) be used in a safety‑critical environment. That story has resonated widely (it had already garnered over 40,000 views on Linkedin) because it illustrates a wider pattern.

Powerful tools are being deployed as if they were just another productivity app, without the basics of change management, risk assessment or support.

In response, I set out a more responsible approach to rolling out tools like Copilot including:

• national guardrails and minimum safety expectations
• local risk assessments aligned with existing patient safety frameworks
• and role‑based training that focuses on how people’s day‑to‑day work could change.

Patients at the centre of digital design

Alongside these technical debates, a recurring question this year has been whether digital tools actually make care safer and easier for patients and staff, or simply add friction.

Articles on the hub have highlighted the importance of co‑design, patient voice and usability – from resources on putting patients at the heart of digital health[10], to reflections on how poor user experience quietly undermines safety.[11]

Wearable AI has been one of the more hopeful technological areas this year, with content[12] showing how real‑time analytics at the wrist[13] or bedside could support earlier detection of deterioration and move care closer to home. This very much aligns with the core shifts described in the new 10 Year Health Plan for England,[14] published in July 2025.

But even with wearables, the same concerns return:

• data overload
• unclear liability when automated insights are missed
• and the need to align new tools with existing (and in many ways, outdated) safety and regulatory frameworks rather than building parallel assurance processes.

Governance, culture and skills for trusted AI

A strong consensus is emerging that trusted AI in healthcare rests as much on people, governance and culture as on technology. Throughout 2025, Patient Safety Learning has highlighted work on transparency and AI governance from organisations such as the Institute for Healthcare Improvement.[15]  

Staff and patients need clarity about data provenance, model limitations and escalation routes if they are to rely on AI‑supported decisions.

Training and culture have been major themes in some of the AI-related talks and workshops I have run, including with colleagues in mental health and addiction services. The focus has been on practical skills – helping people understand when to trust AI, when to challenge it, and how to report concerns. It’s also been valuable to look at building digital safety into everyday practices.

From how PSIRF investigations explicitly include digital and AI systems, through to treating cyber resilience and ‘fragile IT'[16] incidents as core patient safety and business‑continuity risks[17] rather than just technical problems.

Looking ahead to 2026

Stepping back from EPRs, wearables, Copilot and early warning systems, a clear picture emerges from 2025: digital transformation is now inseparable from patient (and clinical) safety. Every new deployment – from a national AI system to a ‘simple’ productivity tool – is a patient safety decision, not just a technology decision.

Looking to 2026, three priorities stand out for anyone working at the intersection of digital health innovation and patient safety:

• Embed safety system learning approaches into digital health change programmes so incidents involving software, data and AI are captured, investigated using models being adopted by PSIRF and shared. And are not treated as IT ‘glitches’.
• Test AI and digital tools against real safety outcomes (whether they actually reduce harm and make care safer for patients), not just efficiency or adoption metrics. be willing to pause or roll back and improve deployments that negatively impact patient safety.
• Invest in people – patients, clinicians, managers and digital teams – so they have the skills, confidence and support to question the tools they use and to shape how technology is introduced into care. Linked to this is the need to focus more on human factors as a core skill within NHS quality improvement teams.

My challenge to all digital health innovators and vendors is:

Treat every digital health project as a patient safety improvement project, and bring patients, staff and ecosystem partners into the conversation early. The pace of change in technology will keep moving quickly; our responsibility is to ensure that safety, learning and trust move just as fast.

Related reading

Electronic patient record systems: Putting patient safety at the heart of implementation (Patient Safety Learning, 31 July 2024)

Patient safety and the role of AI in a cautiously optimistic future: A blog by Ian Fearnley

Balancing promise and risk: AI hallucinations, confabulations and omissions in healthcare

One size does not fit all. How AI and better data can help us embrace complexity in diagnosis and treatment

AI in healthcare translation: balancing risk with opportunity

References

[1] WHO. Artificial intelligence is reshaping health systems: state of readiness across the WHO European Region. 19.11.25.

[2] The Safety Guru with Eric Michrowski. Bridging the Gap: Safety Principles from Aviation to Healthcare Safety with Niall Downey. 22.08.24.

[3] HSJ. Maternity units disrupted for nine months by IT fault. 3.10.2025. 

[4] Digital Health. NHSE warns that AI translation apps could impact patient safety. 3.06.25.

[6] HSSIB. Patient safety issues associated with electronic patient record (EPR) systems – a thematic review. 27.11.25.

[7] NHS England. Identified safety risks with the Euroking maternity information system. 7.12.23.

[8] Flashman, C. LinkedIn Post. 2025.

[9] Flashman, C. LinkedIn Post. 2025. 

[10] Flashman, C. Putting patients at the heart of digital health. Patient Safety Learning's the hub. 14.09.23.

[11] NHS Providers. Making the most of your electronic patient record system. 19.01.23. 

[12] Mahajan, A., Heydari, K. & Powell, D. Wearable AI to enhance patient safety and clinical decision-making. npj Digit. Med. 8, 176. 22.03.25.

[13] Ma, X., Wang, L., Meng, S. et al. A retrospective cross-sectional study showing wearable smartwatches enhance patient safety and efficiency in the intensive care unit. Commun Med 5, 341. 8.08.25.

[14] Department of Health and Social Care. Fit for the Future: 10 Year Health Plan for England. 3.07.25. 

[15]  Moran B, Weckman A, and Martinez N. Transparency and Training: Keys to Trusted AI in Health Care. Institute for Healthcare Improvement. 25.09.25.   

[16] HSJ. ‘Fragile’ IT blamed for critical incidents and patient harm. 20.01.25. 

[17] HSJ. Critical incident declared after EPR launch. 5.11.25.