Blog Post

Artificial intelligence is entering hospitals, clinics and public health programmes faster than many organisations can create the policies, skills and oversight systems needed to manage it responsibly.

The Digital Medicine Society, widely known as DiMe, is attempting to close that gap. As reported by [Healthcare IT News] DiMe has launched a multi-stakeholder initiative to develop practical tools that health systems can use to evaluate, govern, monitor and integrate artificial intelligence.

The announcement is important because healthcare does not lack AI principles. Governments, international organisations, companies and professional bodies have already produced numerous ethical frameworks. What many healthcare organisations still lack are usable processes: procurement checklists, risk-assessment templates, monitoring dashboards, accountability structures and clear procedures for responding when an AI system performs poorly.

DiMe’s new initiative is designed to move healthcare AI governance from policy documents into everyday clinical and operational workflows.

What Has DiMe Announced?

The initiative, called Operationalizing AI Governance in Healthcare, is being developed through DiMe’s Connected Health Collaborative Community.

According to the [official DiMe initiative page] participating organisations will work through focused development sprints to create an open-source toolkit containing templates, scorecards, dashboards and workflows.

These resources are intended to help healthcare organisations answer practical questions such as:

• Who should approve an AI system before it is introduced?
• What clinical, financial, privacy and equity evidence should a vendor provide?
• How should performance be monitored after deployment?
• What happens when a model’s accuracy deteriorates?
• Who is responsible for investigating an AI-related incident?
• When should an AI tool be suspended, retrained or removed?

Qualified Health is sponsoring the initiative, while approximately 30 healthcare, technology, regulatory and professional organisations are contributing expertise. Participants identified in the launch coverage include the US Food and Drug Administration, the American Psychological Association, Intel, Mass General Brigham, Stanford Health, UPMC, Children’s Nebraska, Ferrum Health, Credo AI, the National Health Council and the Consumer Technology Association.

DiMe has said that partners will be active contributors rather than organisations brought in only to review a finished product. Initial resources are expected within two months of the July 2026 announcement, with the complete suite targeted for the end of 2026.

What Effective Operational AI Governance Tools Should Include

DiMe has not yet released the complete toolkit, but its announced focus on templates, scorecards, dashboards and workflows points towards several essential capabilities.

1. AI inventory and risk classification

A health system cannot govern AI tools it does not know it is using. Organisations need a central inventory covering clinical decision-support systems, diagnostic tools, chatbots, documentation assistants, scheduling algorithms, fraud-detection systems and AI features embedded inside existing software.

Each system should then be classified according to its potential impact. An administrative tool that organises meeting notes should not be governed in exactly the same way as an algorithm that influences cancer referrals or emergency triage.

2. Evidence-based procurement

Procurement teams should require vendors to disclose the system’s intended use, training data, validation methods, known limitations, update processes, cybersecurity protections and performance across relevant population groups.

This is especially important for tools imported into African healthcare systems. A model trained primarily on data from North America or Europe may perform differently when disease prevalence, clinical practice, equipment, language or patient characteristics change.

3. Continuous performance monitoring

An AI model’s performance at launch does not guarantee continued performance.

Clinical practice can change. Patient populations can shift. Software integrations may be updated. New equipment may produce different data. These changes can cause model drift, meaning an AI system gradually becomes less accurate or less reliable.

Operational governance therefore requires dashboards that monitor accuracy, false-positive rates, false-negative rates, workflow interruptions, subgroup performance and user-reported incidents over time.

The Role of NGOs and Funding Organisations

NGOs and donors frequently fund AI pilots but may give less attention to the governance infrastructure required after a pilot ends.

Funding agreements should include resources for:

• Independent evaluation
• Clinical workflow redesign
• Community consultation
• Cybersecurity and data protection
• Staff training
• Post-deployment monitoring
• Incident investigation
• Long-term maintenance

Funders should also support African-led research into AI performance, bias and implementation. Evidence generated in well-resourced hospitals cannot automatically be assumed to apply in rural clinics, humanitarian settings or community health programmes.

Governance should therefore be treated as part of implementation, not as an administrative cost added after the technology has been selected.

What Commercial AI Developers Must Provide

Technology companies have an important role in making governance possible.

Vendors should provide clear evidence about intended use, training data, validation populations, excluded groups, performance limitations, model updates and known failure modes. Buyers should also be informed when third-party models are integrated into a product.

Commercial providers should make it technically possible for health systems to monitor performance, export audit logs and investigate incidents. A system that cannot be independently evaluated creates long-term dependence on the vendor.

Research and Funding Priorities

The effectiveness of DiMe’s toolkit will depend on whether it can be tested in different types of healthcare organisations.

Important research priorities include evaluating whether the tools reduce procurement mistakes, improve incident detection, strengthen patient trust and help organisations discontinue unsafe or ineffective systems.

Researchers should also examine whether governance requirements unintentionally disadvantage smaller innovators. Complicated compliance processes can sometimes strengthen large vendors that can afford extensive legal teams while excluding locally developed solutions.

The goal should not be weak governance. It should be proportionate governance: requirements that become more rigorous as the potential risk to patients increases.

Funding should support open, reusable infrastructure, including model documentation templates, shared testing environments, African-language evaluation datasets and independent monitoring systems.

A Practical Starting Point for Health Organisations

Health systems do not need to wait for the full DiMe toolkit before taking action. A practical first phase can be completed by answering five questions:

1. What AI systems are currently in use?
2. Who approved each system?
3. What evidence supports its use in the intended population?
4. How is performance being monitored?
5. Who can intervene when something goes wrong?

Organisations that cannot answer these questions have identified their first governance priorities.

A multidisciplinary AI governance group should include clinical staff, patients, technology teams, legal and privacy specialists, procurement officers, operational leaders and representatives who understand health equity. Smaller organisations may establish a shared regional committee rather than attempting to maintain every capability internally.

Governance Should Enable Responsible Adoption

The fact that DiMe launches an initiative to build operational AI governance tools reflects an important change in the healthcare AI conversation.

The debate is moving beyond whether organisations should have ethical principles. The more urgent question is how those principles influence procurement, clinical workflows, monitoring, accountability and patient protection every day.

For African health systems and the wider Global South, practical governance can be an enabler rather than a barrier. Shared templates and open tools can lower implementation costs, improve purchasing decisions and help health systems identify unsafe technologies before they cause widespread harm.

The success of DiMe’s initiative should ultimately be measured not by the number of documents it produces, but by whether healthcare organisations can use its resources to make better decisions.

Operational AI governance should help health systems adopt valuable technologies with confidence, monitor them continuously, respond quickly to problems and ensure that innovation produces safer, fairer and more accessible healthcare.