AIHE AWARE Healthcare AI Governance Framework

How to Use the New AIHE AWARE Healthcare AI Governance Framework™

Facebook
X
LinkedIn

The hardest part of AI governance begins after approval

Healthcare organisations are getting better at asking some key questions before they introduce artificial intelligence.

Is the technology accurate?
Is it secure?
Has privacy been considered?
Does it comply with regulation?
Has procurement done its due diligence?

All important questions.

But some of the most difficult governance questions arise after the organisation has said yes.

Six months later, is the technology still being used for the purpose that was originally approved?

Has the workflow changed around it?

Are clinicians correcting errors that never reach an incident reporting system?

Has a vendor updated the model?

Are people beginning to trust its recommendations differently?

Has a tool initially introduced for documentation started influencing clinical decisions?

And perhaps most importantly:

Who would know?

This is the problem the AIHE AWARE Healthcare AI Governance Framework™ was designed to address.

AWARE™ is built around five interconnected governance domains:

A — Accountability
W — Workforce
A — Assurance
R — Risk
E — Ethics and Trust

They are deliberately not sequential steps. They are governance conditions that should remain visible throughout the life of an AI system.

The framework places AI that materially influences care within clinical governance while connecting it with digital, cyber, privacy, workforce, legal, procurement and enterprise risk functions.

AWARE is not intended simply to help an organisation approve AI. It’s designed to help leaders remain aware of what happens after approval.

Explore the AIHE AWARE Healthcare AI Governance Framework™

The governance problem is bigger than any algorithm. But much of the conversation about responsible AI understandably concentrates on the technology itself.

Accuracy. Bias. Validation. Cybersecurity. Explainability.

But the problem is that healthcare doesn’t use technology in a laboratory environment. It uses it inside busy emergency departments, outpatient clinics, operating theatres, wards and corporate systems. People interpret it, ignore it, work around it, become dependent on it, challenge it and occasionally use it in ways nobody anticipated when it was purchased.

That means the performance of an AI system cannot be separated from the organisation around it.

A highly capable tool can still become unsafe when accountability is fragmented. An accurate clinical tool can create risk when users do not understand its limitations. A technically validated model can perform differently when the population, data or workflow changes. A tool that promises five hours of productivity may simply transfer five hours of checking, correction or administration somewhere else. And an AI application may continue operating long after the evidence supporting its value has weakened.

The AWARE Framework therefore starts from a broader proposition:

Healthcare AI governance is not simply about governing technology.

It’s about governing the relationship between technology, people, clinical care and organisational responsibility.

This is also why AIHE argues that AI influencing patient care should not sit inside a separate technological governance universe. Clinical governance already exists to oversee safety, quality, clinical effectiveness, workforce capability, consumer participation, risk and improvement. AI increasingly touches every one of those domains.

The challenge is to connect them.

So how should an organisation actually use AWARE?

The easiest mistake would be to turn AWARE into another checklist completed before implementation. That would miss much of its value.

A better starting point is to take one AI application currently being used or seriously considered within the organisation and put it through all five AWARE lenses.

Imagine, for example, an AI documentation system. At first glance, it appears relatively straightforward. It listens to a consultation and generates a clinical note. But look at the same technology through AWARE and the governance conversation quickly becomes richer.

A — Accountability: who owns the consequences?

The first question is deceptively simple: Who is accountable for this AI system?

Not who bought it. Not who manages the contract. But, who has sufficient authority to obtain information, investigate concerns, require changes and, if necessary, stop its use?

AWARE recommends clear executive and clinical ownership, supported by defined reporting, escalation and decision-making authority. Accountability should extend from Board oversight through executive sponsorship to clinical, operational and technical responsibilities.

For the documentation tool, that might mean asking:

  1. Who monitors its clinical performance?
  2. Who investigates repeated inaccuracies?
  3. Who decides whether it can be used for new purposes?
  4. Who is informed when the vendor changes the underlying technology?
  5. Who has authority to suspend its use?

These questions become particularly important because the phrase “the clinician remains accountable” can inadvertently become an organisational escape hatch.

Clinicians certainly retain professional responsibilities when using AI. But organisations also have responsibilities for the conditions under which clinicians are expected to use it: selection, implementation, training, monitoring, escalation and governance.

Human accountability does not mean organisational accountability disappears.

W — Workforce: can our people actually use it safely?

The second question shifts attention from the software to the people using it. Are our people capable of using AI safely?

Training someone to click the correct buttons is not the same as preparing them to work safely with AI.

Clinicians may need to understand when outputs should be challenged, what limitations exist, how verification should occur, what should be documented and how concerns should be escalated.

Executives need different capability: governance literacy, risk interpretation and an understanding of what evidence should constitute meaningful assurance.

Boards need enough AI governance literacy to ask informed questions without needing to become technologists.

The practical question is therefore not “Has everyone completed the training module?” It is “Can the people using this system recognise when it may be wrong — and do they know what to do next?”

That is a much higher standard.

And it raises another uncomfortable question. Does the workflow actually give clinicians enough time to perform the checking that governance expects of them?

Because a policy requiring human verification means very little if operational conditions make meaningful verification unrealistic.

A — Assurance: how do we know our confidence remains justified?

This may be the most important shift in the framework.

Implementation is not evidence of success. Usage is not evidence of safety. And absence of reported incidents is not necessarily evidence that nothing is going wrong. AWARE treats assurance as an active evidence-generating function.

The framework identifies eight areas of assurance: safety, quality, equity, value and financial performance, workforce adoption, consumer trust, cybersecurity and compliance.

For an AI documentation tool, assurance might therefore look beyond whether the system is available and being used.

  1. Are correction rates increasing?
  2. Are particular types of information repeatedly omitted?
  3. Are clinicians creating workarounds?
  4. Has documentation quality improved?
  5. Has workload genuinely fallen?
  6. Are complaints appearing?
  7. Are some specialties or patient populations experiencing different performance?
  8. Are clinicians becoming overly reliant on the generated record?

Those are all governance signals. In fact, some of the most valuable indicators may emerge before anybody is harmed. A rise in overrides. More manual corrections. Growing workarounds. A sudden change in user confidence. Unexpected variation between departments. These are not merely implementation inconveniences. They may be early warning signals.

Mature governance pays attention while there is still time to act.

R — Risk: not every AI system should be governed the same way

Healthcare could easily respond to uncertainty about AI by subjecting every application to the same complex governance process.

That would create bureaucracy without necessarily creating safety. AWARE instead adopts proportionate governance.

The framework describes AI applications across low, moderate, high and critical risk levels, with greater oversight as the potential consequences for patients and organisations increase. An administrative tool with limited consequence if incorrect should not necessarily undergo the same process as technology influencing diagnosis, treatment or prioritisation.

This is where AWARE can become particularly useful for executive teams.

The question becomes: What could happen if this technology is wrong, unavailable, misunderstood, misused or used outside its intended purpose?

That assessment should consider far more than clinical accuracy. Privacy, cybersecurity, automation bias, misinformation, discrimination, vendor dependency, business continuity, data quality, legal ambiguity and reputational consequences may all matter.

Governance intensity can then follow the risk rather than the excitement surrounding the technology.

E — Ethics and Trust: would patients be comfortable if they understood exactly how we were using it?

The final AWARE question is one healthcare leaders may find particularly powerful:

Would patients trust how we are using AI?

Not simply whether its use is technically legal. Whether it is fair, transparent, respectful of autonomy and consistent with the kind of healthcare relationship the organisation claims to provide.

The framework asks organisations to consider transparency, explainability where possible, bias, privacy, inclusion, patient autonomy and preservation of appropriate human judgement and connection. This matters because AI is creating forms of organisational capability that healthcare has not previously possessed.

A documentation system can become a productivity surveillance system. Decision-support technology can create a record of every occasion on which a clinician disagreed with an algorithm. Workforce optimisation can become workforce monitoring. Predictive systems may influence access to care before a patient even encounters a clinician.

The governance question therefore changes from “Can we do this?” to “What obligations arise because we now can?

Technological capability does not automatically create legitimacy. Trust has to be earned.

Use AWARE before implementation — but also use it on AI you already have

One of the most practical uses of the framework may be retrospective.

Many health services already have AI-enabled technology embedded across clinical and corporate systems. Some organisations may not even have complete visibility of where it exists.

AWARE can therefore be applied as a portfolio review.

Start by establishing an inventory of AI-enabled systems. Identify their intended purpose, executive and clinical owners, affected workflows and level of risk. Then examine each system against the five AWARE domains and identify where evidence is strong, weak or absent.

The framework’s implementation pathway recommends establishing executive and Board accountability, creating an AI inventory, undertaking proportionate risk assessments, building workforce capability, introducing ongoing assurance and reporting regularly through governance structures.

Importantly, this does not require a hospital to build an entirely new bureaucracy.

AWARE is designed to connect existing clinical governance, executive, risk, digital, privacy, workforce and Board systems. An AI governance committee may be useful as a coordinating function, but the framework cautions against turning it into another silo. Clinical safety should remain connected to clinical governance; cyber risk to cybersecurity governance; privacy to information governance; workforce capability to workforce systems.

The objective is integration.

Not another committee that receives another report.

The most important question here may be “What would make us reconsider?”

Healthcare governance has traditionally been good at approving things. It has sometimes been less disciplined about asking when an approval should cease to apply. AI makes that question increasingly important.

The AWARE continuous assurance cycle asks organisations to understand, measure, monitor, review and improve AI throughout its life. Material changes — including model updates, changes in data, new populations, altered workflows, incidents or unusual output patterns — should prompt reassessment.

That means every significant AI deployment should have an answer to another question:

What evidence would cause us to restrict, redesign, suspend or retire this system?

That question changes governance considerably. Because if an organisation knows what evidence would make it stop, it also has to decide what evidence it is going to collect. And that is where assurance becomes meaningful. Continuing to use AI because nobody has formally decided to stop is very different from continuing to use AI because current evidence supports that decision.

What should a Board see?

Boards do not need dashboards containing every metric associated with every AI application. They need meaningful organisational assurance.

A Board using AWARE should increasingly be able to see the organisation’s AI portfolio, high and critical risks, material incidents, unresolved assurance gaps, important trends and overall governance maturity. Executives require greater operational visibility into ownership, risk profiles, value and performance; clinical governance requires visibility of validation, incidents, overrides, outcome variation, equity and revalidation.

That allows the Board conversation to move away from “How many AI projects have we implemented?”, towards “What evidence gives us confidence that the AI operating in this organisation remains safe, valuable and worthy of trust?”

That is a much more useful governance question.

From AI adoption to AI governance capability

There will be considerable pressure over the next few years for healthcare organisations to demonstrate that they are adopting AI. That may be the wrong measure of maturity.

The AIHE AWARE Framework describes governance maturity from ad hoc and fragmented arrangements through to integrated systems capable of continuous assurance, review and improvement. Its central proposition is that maturity should not be measured by how much AI an organisation has deployed, but by its ability to understand, govern, monitor and, when required, change or stop it.

This is an important distinction. Because a hospital with fifty AI applications and limited organisational visibility may be less mature than one with five systems that are clearly owned, appropriately governed and continuously assured. The goal is not to become AI-heavy. The goal is to become sufficiently governance-capable to use AI confidently at the level of risk the organisation chooses to carry.

The AWARE question to take back to your organisation

There is a simple way to begin.

Choose one AI system already operating in your organisation.

Then ask five questions:

  1. Who is accountable for it?
  2. Are our people capable of using it safely?
  3. What evidence tells us it continues to perform as intended?
  4. Are its risks understood and governed proportionately?
  5. Would patients trust the way we are using it?

If those questions produce five confident answers, the organisation has a strong starting point. If they reveal uncertainty, fragmented responsibility or an absence of evidence, that is useful too. Because good governance begins by making the invisible visible.

Healthcare will not become safer simply because its technology becomes more intelligent. As technology becomes more capable, our systems of human accountability need to become clearer.

That is the purpose of AWARE™: not to slow responsible innovation, but to help healthcare organisations build the governance confidence to lead it well.

Explore and download the AIHE AWARE Healthcare AI Governance Framework™

For further practical tools, guidance and resources on AI governance, visit the AIHE AI Governance Knowledge Hub.