AI in Healthcare 2027: What Clinics Can Use Safely and What to Ask Vendors

AI in healthcare 2027 will mean different things to different clinics, from note-drafting tools and chat assistants to software that flags patients at risk, and the sensible question for a clinic is not which tool is newest but which ones can be used safely. Safe use depends on what the tool does, whether it counts as regulated medical software, what data it sees and who checks its output. This guide summarises what the World Health Organization, the United States Food and Drug Administration and the United States Office of the National Coordinator for Health IT say on their own pages, read on 17 November 2026, and turns it into questions a clinic can put to any vendor. It does not predict which products will work, and several detailed criteria could not be read, which the guide notes. It is general information, not legal or medical advice.
Key takeaways
- The WHO's guidance on ethics and governance of artificial intelligence for health sets out six consensus principles and says AI for health must put ethics and human rights at the centre of design, deployment and use.
- WHO's 2025 guidance on large multi-modal models describes them as general-purpose foundation models and notes it is not yet proven they can handle a wide range of tasks.
- The FDA's January 2026 final guidance on clinical decision support software explains which decision support functions are excluded from the device definition under the Cures Act, and says its digital health policies still apply to software that meets the device definition.
- The FDA says its list of AI-enabled devices is not comprehensive, and that listed devices met the agency's applicable premarket requirements.
- The HTI-1 rule sets transparency requirements for predictive algorithms in certified health IT, so clinics should ask vendors what information they can provide about an algorithm.
AI in healthcare 2027: what kinds of AI clinics are being offered
Clinics are being offered several different things under one label. Some tools draft clinical notes or letters from a conversation, some answer patient messages, some summarise records, some help with scheduling and reminders, and some claim to flag diagnoses or risks. These carry very different levels of risk. A tool that suggests a reply to a patient's booking question is a different matter from one that suggests a diagnosis. The first step for any clinic is to write down exactly what the tool would do, what data it would see and what decision, if any, it could influence.
What the WHO says
The WHO publication Ethics and governance of artificial intelligence for health, dated 28 June 2021, says it is the product of eighteen months of expert deliberation, identifies ethical challenges and risks and sets out six consensus principles to guide the design, development and use of AI for health. It says ethics and human rights must be at the heart of design, deployment and use, and it gives governance recommendations that hold public and private stakeholders accountable. A later WHO guidance document on large multi-modal models, dated 25 March 2025, describes these models as general-purpose foundation models that can take several types of input and produce outputs not limited to the input type, and says it is not yet proven that they can perform a wide range of tasks reliably. The landing pages read do not list the individual principles, so clinics should read the full guidance.
When software becomes a regulated medical device (United States)
The FDA's final guidance on clinical decision support software, dated January 2026 and issued jointly by its device, biologics and drug centres, says the Cures Act amended the Federal Food, Drug, and Cosmetic Act to exclude certain decision support software from the device definition, and that the guidance clarifies which functions meet the non-device criteria. It also says the FDA's digital health policies continue to apply to software functions that meet the device definition. The detailed criteria were in a document that could not be read for this guide, so they are not listed here. The FDA also maintains a list of AI-enabled medical devices authorised for marketing, which it says is not a comprehensive resource and is updated periodically. For a clinic, the practical point is to ask a vendor whether the product is regulated as a medical device in the clinic's country and, if the vendor says it is not, why.
Algorithm transparency in certified health IT
The Office of the National Coordinator's page on the HTI-1 rule, last updated on 7 November 2025, says the rule sets transparency requirements for artificial intelligence and other predictive algorithms that are part of certified health IT, so that clinical users can access a consistent baseline of information about them and assess fairness, appropriateness, validity, effectiveness and safety. The fact sheet with the specific requirements could not be read, so details and dates are not stated here. Clinics outside the United States are not bound by this rule, but its logic is useful anywhere: a vendor should be able to say what an algorithm was trained for, how it was tested and what its limits are. The guide to healthcare software trends in 2027 for the USA covers the related interoperability rules.
Questions to ask every AI vendor
- What exactly does it do, and what decision could its output influence?
- Is it regulated as a medical device in my country? If not, on what basis?
- What data does it see, where is that data processed and stored and is it used to train models?
- Who reviews the output before it enters a record or reaches a patient?
- What happens when it is wrong: how are errors reported, logged and fixed?
- What evidence exists, such as testing on patients like mine, and who produced it?
- How can I switch it off, export my data and stop using it without losing records?
Safe starting points for a small clinic
- Administrative tasks first: reminders, scheduling and drafting non-clinical messages carry lower risk than clinical advice.
- Human review always: a clinician should read and sign off anything that goes into a patient record.
- Minimise data: give a tool only what it needs and avoid identifying details where possible.
- Pilot with a log: test on a small group, record errors and decide with evidence.
- Tell patients: explain when AI is used and give a way to opt out where appropriate.
- Check local law: privacy, health data and device rules differ by country.
Where CreativeIn Tek fits
CreativeIn Tek builds AI agents through its AI solutions service and the AI Agent product, which answers customer messages on WhatsApp, linked by QR code. It does not provide diagnosis or clinical decision support, and nothing in its products is certified as a medical device. CareTrack, its clinic management system, has no AI diagnosis features. The difference between a chatbot and an agent is explained in the guide to AI agents versus chatbots, and a clinic can talk to CreativeIn Tek about administrative automation.
Frequently asked questions
Is AI safe to use in a clinic?
It depends on the task. Administrative uses carry lower risk than clinical advice, and any output that reaches a record or patient should be reviewed by a clinician.
Is clinical AI software regulated?
Some is. The FDA says certain decision support software is excluded from the device definition and other software that meets the definition remains regulated. Ask each vendor about the rules in your country.
What does the WHO recommend?
The WHO guidance says AI for health must put ethics and human rights at the centre and sets six consensus principles, and its 2025 guidance says it is not yet proven that large multi-modal models can handle a wide range of tasks.
What is algorithm transparency?
In the US HTI-1 rule it means certified health IT must give clinical users a baseline of information about predictive algorithms so they can assess fairness, validity, effectiveness and safety.
Can AI write patient notes?
Some tools draft notes, but a clinician should review and approve them, and the clinic should check where the data goes and how it is stored.
Last reviewed: 17 November 2026, against the WHO guidance on ethics and governance of artificial intelligence for health and on large multi-modal models, the FDA clinical decision support software guidance and the HTI-1 rule page. The detailed FDA criteria and HTI-1 requirements were not read. Rules change, and this article is general information and not legal or medical advice.


