The SignalNo. 1Sample issue
Sample issue. This shows the shape of The Signal: short notes on what changed and what it means for a clinic or health system, grouped by area, each with its source. The items below are standing references rather than this week's news, so the format can be reviewed without inventing events.
Before a vendor conversation about any AI tool that touches diagnosis or treatment, check whether the product, or the model inside it, appears on the FDA's list of AI-enabled devices. If it does not, ask the vendor why, and in writing. The list is updated periodically and the count changes, so cite the date you looked.
U.S. Food and Drug Administration(opens in a new tab)The AMA's 2026 physician survey reports that 72% of physicians use AI in at least one part of their practice, up from 38% in 2023, and that 92% want more education and training on it. For leaders, the second number is the actionable one: adoption is already happening; preparation is what lags.
American Medical Association, 2026 Physician Survey on Augmented Intelligence(opens in a new tab)OWASP's list of the top risks for applications built on large language models puts prompt injection first: text in a document or a message that the model mistakes for an instruction. A faxed referral or a patient message can carry that text. A vendor who cannot explain their defense in plain terms has not thought about your inputs.
OWASP Foundation(opens in a new tab)HHS guidance explains when a vendor that creates, receives, maintains or transmits protected health information on your behalf is a business associate, and what agreement that requires. Settle this before a pilot, including for tools that only read policies, because staff questions to those tools often contain patient details.
U.S. Department of Health and Human Services, Office for Civil Rights(opens in a new tab)The ONC's HTI-1 final rule requires certified health IT developers to disclose 'source attributes' for predictive decision-support interventions: what data trained a tool, how it was tested, and how it is monitored. Most clinic automation sits outside the rule, but the expectation is a sensible one to bring to every vendor.
Office of the National Coordinator for Health IT(opens in a new tab)If you are writing or revising an internal AI policy, the NIST framework and its generative AI profile give you four plain functions to organize it around: govern, map, measure, manage. You do not need to adopt all of it. Borrowing the structure saves arguing about one from scratch.
National Institute of Standards and Technology(opens in a new tab)This is a sample issue. It exists so the format of The Signal can be reviewed in the admin and on the public page before the first real issue is written.
A real issue follows the same shape: one short note per item on what changed, one or two sentences on what it means for a clinic or health system, and the source. Items are grouped by area so a reader can skip to what they own. Nothing is included without a source a reader can open.
One Friday email on clinical AI for clinic and health system leaders. What changed, what it means, and what to ignore.