Welcome to this week’s Dental AI Weekly, honest analysis of where dental AI is going, from someone building in it.
A dental AI demo can look great in a quiet room. A full schedule asks harder questions. Does it work on the patients and problems in front of you? Does the evidence match the claim? Who notices when the system gets something wrong?
Three developments this week moved those questions closer to the center of healthcare AI. Saudi Arabia authorized a locally developed dental X-ray system. Researchers compared six AI systems on restorative-dentistry questions. A new federal program began funding clinical AI designed to take action in cardiovascular care.
The adoption race is becoming an evidence race. A strong answer is only the start. Clinical AI also has to prove where it works, show where it does not, and keep earning trust after launch.
WHAT HAPPENED THIS WEEK
01 – X-RAY REVIEW
Saudi Arabia authorized Dental IQ, a locally developed AI system that flags possible findings on dental X-rays.
The Saudi Food and Drug Authority granted marketing authorization to Dental IQ on September 10. The system uses software trained to recognize patterns in dental X-rays and can flag possible caries, periodontal disease, and structural changes affecting teeth and supporting tissues.
The authority says its review included technical documentation, clinical evidence, local clinical studies, and plans for follow-up after launch. It did not publish the study size or performance figures. The clinician remains responsible for reviewing the output and making the clinical decision.
What this means: Authorization tells a dentist that a regulatory standard was met in one market. Published performance data and follow-up tell us how much confidence the result deserves across different patients, images, and care settings. Clinical AI will increasingly be judged on both.
02 – RESTORATIVE BENCHMARK
A BMC Oral Health study found meaningful performance differences among six AI systems answering restorative-dentistry questions.
Researchers tested six current AI systems on 195 text-based questions from Turkey’s Dentistry Specialization Examination, covering exam years 2012 through 2025. Overall accuracy ranged from 83.6% to 96.4%, and the differences among systems were statistically significant.
The authors concluded that the systems may be useful as complementary resources for dental education and exam preparation. They also made the boundary clear: correct answers on an examination do not establish educational effectiveness, reasoning ability, or clinical safety.
What this means: The adoption race is becoming an evidence race. “AI” is not one level of performance, even when every system gets a strong score. The spread between six products on the same dental questions is a reminder to ask for evidence from the exact task a tool is supposed to handle. An exam benchmark is useful evidence, but it is not a patient-care trial.
03 – CLINICAL AGENTS
ARPA-H awarded teams to build and test a clinical AI system intended to seek FDA authorization for cardiovascular care.
The Advanced Research Projects Agency for Health announced the selected teams for its four-year ADVOCATE program on September 9. The program’s goal is a patient-facing clinical agentic AI system, meaning software designed to take multiple steps toward a care goal rather than answer a single prompt. ARPA-H committed up to $33.7 million in the first year of the $62.7 million program.
The teams are expected to work with the FDA, submit an authorization package within 24 months, build continuous safety monitoring, and test the system across health systems and rural sites. Johns Hopkins University Applied Physics Laboratory will provide independent evaluation. This is a development program for cardiovascular care, not an authorized product available today.
What this could mean for dentistry, OraCore editorial analysis: The program treats clinical AI as more than a model. Regulation, independent testing, workflow integration, ongoing monitoring, and payment all have to develop around it. Dentistry will not inherit this cardiovascular program, but it may inherit the standard patients and regulators come to expect from AI that takes action in care.
“The adoption race is becoming an evidence race.”
THE BIGGER SIGNAL
These three developments put evidence at different points in the product’s life. Dental IQ reached authorization with local studies and a plan to watch performance after launch. The restorative benchmark showed that six capable systems still produced different results on the same set of questions. ADVOCATE is building independent evaluation and continuous monitoring into a clinical agent before it reaches patients.
That changes what a credible AI claim looks like. A vendor still needs to show that the software can do the task. The next questions are where the evidence came from, whether it matches the intended use, and what happens when performance slips in the real world.
For dentists, the useful divide is getting clearer. A benchmark can show capability. A clinical study can test use in a care setting. Authorization can permit a product in one market. Monitoring can show whether it keeps behaving after launch. None replaces the others.
BY THE NUMBERS
66%
of nearly 1,200 physicians surveyed by the American Medical Association reported using health AI for at least one of 15 tasks in 2024, up from 38% in 2023.
What it signals: Self-reported use grew quickly in medicine, but the survey did not measure whether those tools stayed in daily use or improved care. Access and lasting adoption are different results.
Source: American Medical Association, February 26, 2025.
READER Q&A
“We tried an AI tool last year, but the team stopped using it after two weeks. What should we do differently before trying again?”
Practice question selected for Issue 030
BH: This is common, and it is not unique to AI. Practice owners are often surprised when they audit their software and find tools that became obsolete, were forgotten, or simply never earned their place, yet keep hitting the bottom line every month.
AI should be evaluated the same way. Start with what the practice actually needs, then keep only the software that supports those needs. Every tool also needs a champion who owns the training, daily adoption, and accountability.
A lot of software fails in a practice because nobody owns the rollout after the purchase. Buying the tool is easy. Building it into the team’s daily workflow is where the result is won or lost.
FROM THE ORACORE BLOG THIS WEEK
Workflow | 3 min
Dental AI Workflow Agents Need More Than Scheduling Access
An AI agent can reach the schedule and still leave the team moving information by hand. This article looks at the clinical context, review controls, ownership, and failure handling needed when software connects several steps.
Try OraCore Scribe for 14 days with no credit card. Choose Scribe Solo for one provider or Scribe Team for a full practice workflow.
That’s the week. Reply with what you’re seeing in your own practice. I read every one.
Brad Hutchison
CEO, OraCore AI
Dental AI Weekly drops every Monday. pulse@oracorenews.com | OraCore AI, Denver CO