AI in healthcare isn’t a someday conversation anymore. It’s already in the building, and mostly, it’s doing paperwork.
The American Medical Association’s 2026 Physician Survey on Augmented Intelligence found that 81% of physicians reported awareness or use of AI in their practice, up from 38% in 2023. Narrow that to physicians using at least one specific AI tool and the figure is 72%.
Still, the interesting question isn’t whether healthcare will adopt AI—it’s where AI is earning its keep.
AI has gained ground in the work that eats clinician time but still passes through a person before it reaches a patient record. And because that work runs on protected health information, every gain arrives with a question attached: where does patient data go once it starts moving through AI tools and onto mobile devices?
Read on to explore how healthcare organizations are using AI today, where the technology may go next, and why security belongs in the deployment plan from the start.
Table of contents:
- How AI is already being used in healthcare
- Where AI in healthcare requires more caution
- The future of AI in healthcare—and what has to be true to get there
- Securing AI in healthcare with Samsung Knox
How AI is already being used in healthcare
AI has taken hold wherever a clinician still reviews the output and no clinical decision transfers to the model. Documentation, summarization, and billing lead the way. Anything closer to diagnosis trails behind.
Clinical documentation
Documentation is the most established use case in medical practice. Nearly 4 in 10 physicians in the AMA survey reported using AI for summaries of medical research and standards of care, a use that jumped 26% from 2024. Drafting discharge instructions, care plans, and progress notes follows close behind, along with billing and chart documentation.
What AI delivers here is a better starting point, not a finished product. Instead of building every document from a blank page, a clinician reviews and corrects a draft.
There’s a reason this category leads. Most physicians see AI as a way to automate the tasks that fuel burnout, and documentation is where that pressure shows up first.
Operational and admin work
Beyond the chart, physicians report using AI for billing code documentation, chart summaries, and drafting replies to patient portal messages. None of it is visible to patients the way a diagnostic tool is. All of it is work that quietly piles up across a practice.
By clearing avoidable manual effort, clinical and operational staff can spend their time where their expertise actually matters.
Where AI in healthcare requires more caution
Adoption drops off sharply the moment AI touches clinical judgment or speaks directly to patients. These applications carry requirements the admin ones don’t: clinical validation, regulatory oversight, and a clear path to escalate to a person.
Predictive analytics
Predictive models look for patterns in clinical data and surface signals that may deserve a closer look. The FDA notes that AI-enabled medical devices are being developed for uses including prognosis, treatment-response prediction, and risk assessment.
Think of these tools as a way to prioritize where to look inside a complex dataset. The output still needs a clinician to read it against the patient’s history, current condition, and care context.
Medical imaging
AI-enabled imaging software can flag possible findings, prioritize studies for review, and take measurements. To see what has actually cleared review, the FDA publishes a list of AI-enabled medical devices authorized for marketing in the U.S. Devices appear on it after meeting the agency’s premarket requirements, which include a focused review of safety and effectiveness.
It’s tempting to file imaging under “heavily regulated” and move on, but the picture is more specific. The FDA regulates software functions that meet the definition of a medical device, while certain clinical decision-support functions can fall outside that definition when they meet specific criteria.
Patient-facing assistants
A well-scoped virtual assistant can field routine questions, help patients navigate scheduling and intake, and draft responses for staff to review, with a clear path to a person the moment something turns sensitive or clinically significant. Increasingly, that same idea is showing up outside the clinic, helping people understand their own health day to day instead of waiting for an appointment to ask about it.
Samsung Health Assistant, in beta for eligible U.S. users, is one example: it connects a person’s sleep, activity, nutrition, mindfulness, and vitals data and turns it into insights they can act on, surfacing patterns they might otherwise miss. That’s the same shift AI is making everywhere else in this piece, applied to the person instead of the practice—less time spent piecing together what the data means equals more time acting on it.
The future of AI in healthcare—and what has to be true to get there
Where AI goes next has less to do with model capability than with a harder question. Can organizations push past that dividing line without giving up safety, privacy, or trust? Three areas will shape the answer:
- Care delivery: AI may help clinicians pull information together faster and personalize support, though wider use depends on validation and human oversight.
- Admin relief: Documentation and patient communication will likely stay the headline uses, since neither hands clinical responsibility to an algorithm.
- Drug development: The FDA is examining AI’s use across the drug development lifecycle, from discovery through post-market monitoring.
Trust sets the pace for all of it. Physicians consistently rank safety validation and data privacy assurances as their top requirements for adoption, with clear liability frameworks the regulatory step they ask for most.
One more finding from the same survey is worth pausing on: nearly 1 in 10 physicians weren’t sure which AI tools their own practice offers. In some places, adoption is moving faster than the visibility around it.
So, evaluating a healthcare AI tool takes more than asking whether it works. It means knowing what data the tool receives, how long that data sticks around, who can access it, and which devices are allowed to connect to the workflow at all.
Securing AI in healthcare with Samsung Knox
That last question is where mobile security enters the picture. The Department of Health and Human Services states that regulated entities may use mobile devices to access electronic protected health information when appropriate safeguards are in place. That puts device security inside the AI strategy rather than beside it.
If your organization runs Samsung Galaxy devices, Knox Platform for Enterprise answers it with device health attestation: your IT team can confirm a device is in a trusted state before it connects to a clinical system, instead of trusting that someone is signed in. That’s the same principle a Zero Trust approach runs on, applied at the endpoint.
AI will change how healthcare professionals work. Whether people trust that change comes down to the safeguards built around it.
Ready to build a more secure mobile foundation for healthcare AI?
