What is the NHS’s approach to AI and how does this foreshadow the future of AI governance across all sectors?
In October, the government confirmed it will accept all 44 recommendations made by the National Commission into the Regulation of AI in Healthcare and set out how they will be delivered across the UK.
Alongside the response, the MHRA is opening the call for applications for the third phase of AI Airlock, its regulatory sandbox for AI-enabled medical devices, with testing themes directly informed by the Commission’s recommendations.
The recommendations raise important questions about the processing of health data, transparency to patients, allocation of responsibility between AI developers and healthcare providers, cybersecurity and the governance of AI systems throughout their lifecycle.
These issues are relevant to pharmaceutical and life sciences companies, healthcare providers, and medical technology developers. They are also important for any organisation developing, supplying or deploying AI-enabled healthcare solutions.
And experts believe the recommendations apply beyond the healthcare sector too. If you work in compliance, security, risk, or governance outside healthcare, it is tempting to file these under “someone else’s regulator”, says Tracey Hannan-Jones, consulting director in information security, UBDS Digital.
However, she believes all sectors should be paying close attention. “Healthcare is where the consequences of AI failures are most visible, so it tends to show early where good governance is heading.”
What does the NHS’s approach foreshadow about the future of AI governance more broadly — in particular, the shift from assessing AI before deployment to continuously governing its performance, risks and impact throughout its lifecycle?
Single Point of Assessment
As part of the changes, the UK government is moving away from a single point of assessment to a “continuous lifecycle approach” for AI in healthcare.
Dray Agha, senior manager, tactical response at Huntress explains how the shift acknowledges that AI models evolve as they process new patient data and encounter evolving clinical scenarios. “A static security or safety check is no longer sufficient to protect patient outcomes. We need ongoing monitoring to catch algorithmic drift or new vulnerabilities as they emerge.”
The shift towards a continuous lifecycle approach has practical consequences that apply “well beyond medical devices”, according to Hannan-Jones.
Indeed, across all sectors, if your organisation relies on a point-in-time supplier assurance, an approval at procurement and an annual review, change is needed, she says.
Governance Questions
To address AI governance, the healthcare sector and other relevant organisations should consider:
- Approval is not the same as assurance. An AI system assessed as appropriate today may behave differently as its models, data, environment or intended use change.
- Accountability must extend beyond the technology provider. Organisations deploying AI need to understand their own responsibilities, rather than relying on the supplier’s assurances.
- AI governance cannot operate in isolation. Information security, privacy, operational risk, human oversight and organisational accountability are all involved.
- Evidence matters throughout the lifecycle. Organisations need to demonstrate that controls continue to operate effectively, not merely that an assessment was completed before deployment.
AI as a Dynamic Asset
The NHS appears to treat AI as “a dynamic asset, rather than a static piece of software”, according to Huntress’ Agha.
Agha believes the “continuous monitoring” model sets a precedent that other sectors can follow to manage AI safely. “Any business deploying AI must build the capability to monitor performance and security throughout the entire lifespan of the tool,” he advises.
Tristan Shortland, CTO at Infinity, concurs. He believes AI governance in general is moving away from the idea that “you can assess a system once, approve it and assume the risk remains static”.
“AI systems operate in changing environments, interact with changing data and, in some cases, evolve over time,” Shortland tells IO. “Governance therefore must continue throughout the lifecycle of the technology.”
To prepare for this now, businesses across all sectors should be thinking about what happens after an AI solution goes live, says Shortland. He advises asking: “Who owns it? What is it allowed to access or do? How is its performance monitored? How do you identify unexpected behaviour, changes in output quality or emerging risks? And when does a human need to intervene?”
It becomes even more important as organisations move from relatively contained use cases such as copilots towards AI agents capable of taking actions across business systems, according to Shortland.
Integrate AI Governance
The NHS approach to AI is certainly food for thought across all sectors as the technology becomes part of everyday operations.
As a first key step, security leaders must integrate AI governance into their existing risk management frameworks “immediately”, advises Huntress’ Agha. “You need to map exactly where AI is currently being used across your organisation and establish baseline performance metrics. Security teams should also set up automated monitoring to flag anomalous behaviour in these models over time.”
Leaders should inventory AI systems, assign accountable owners, map data access, and establish risk-based monitoring, adds Dana Simberkoff, chief risk, privacy and information security officer at Avepoint. “Define when changes require reassessment, preserve audit trails, and test escalation and shutdown procedures. Treat governance as an ongoing operational responsibility: Know what AI is doing, who can intervene, and how to demonstrate control.”
After gaining visibility, classify use cases by risk and establish minimum controls, advises Infinity Group’s Shortland. “Define which tools are approved, what data can be used, where human review is mandatory and what needs to be logged and monitored. Security teams should also treat AI identities, agents and integrations with the same discipline they apply to human users, using least-privilege access, rather than providing AI with broad access by default.”
At the same time, leaders should ensure they make governance continuous, rather than a one-off task, recommends Shortland. “Review permissions, monitor outputs and usage, record incidents and reassess higher-risk use cases as the technology, data or business process changes. AI is evolving too quickly for governance to be a once-a-year compliance exercise.”
Standards can give organisations a much more structured way to approach this. These should complement the controls organisations already have around information security and privacy, says Shortland. “ISO/IEC 27001 provides the foundation for managing information-security risk, while ISO/IEC 27701 extends that thinking into privacy information management. ISO/IEC 42001 adds the AI-specific management layer.”
The important point is that AI governance shouldn’t sit in isolation, Shortland concludes. “AI relies on your existing data, identities, systems and security architecture. The strongest approach is therefore to integrate AI governance into the management and control environment you already have, then continually review it as AI adoption expands.”
Expand Your Knowledge
Blog: Healthcare AI Is Moving Fast, But Data Governance Is Struggling To Keep Up
Blog: Data Governance in Healthcare in the Age of Ransomware
Case study: How Healthcare RM is streamlining compliance and delighting auditors with ISMS.online







