AI Regulation Tracker  /  Licensing boards

Washington is reaching Delaware's result on AI clinicians without passing a statute

Two draft documents went to the Washington Medical Commission's Policy Committee on July 22, 2026. Neither is in force. Together they would say that only humans may hold a medical license, and that leaning on AI for clinical judgment can put a license at risk.

By Anthony Guerriero  ·  Published July 28, 2026  ·  Last updated July 28, 2026  ·  Standard news article

What actually happened on July 22?

The Policy Committee agenda for July 22, 2026 lists the interpretive statement at item 2, pages 2 to 4 of the packet, and the AI patient-care policy at item 3, pages 5 to 11. Both are drafts. Taking a document up in committee is not adopting it. Nothing in the packet changes a licensee's obligations today.

The interpretive statement is the shorter of the two. Its stated purpose, in the draft's own words, is "to formally articulate the commission's position that licensure is limited to human individuals, and that non-human entities, technologies, or augmented intelligence systems are not capable of and may not independently hold a medical license or practice medicine as defined by law."

The reasoning is procedural rather than philosophical, which is what gives it teeth. Licensure presupposes a natural person who can be educated, examined, assessed for moral character, disciplined, and afforded due process under the Administrative Procedure Act. A model has none of those attributes. The draft meets the obvious counterargument head on: "While recent developments have shown that augmented intelligence tools have the capability of passing the U.S. Medical Licensing Examination, that is one step in the process of gaining a medical license and demonstrating competent clinical practice."

"The statutory framework leaves no mechanism for non-human entities to obtain licensure, satisfy moral or physical capability requirements, or assume legal responsibility for patient care. Therefore, under Washington law, only humans may lawfully practice medicine."Draft interpretive statement, conclusion, WMC Policy Committee packet, July 22, 2026

Why does the route matter more than the result?

Delaware got to the same place by statute. House Bill 191 barred non-human entities from holding a medical license or using clinician titles, and it went through a legislature and a public record that anyone tracking state AI bills would have caught. Washington is getting there by having a licensing board read an existing statute, RCW 18.71.021, and publish what it thinks that statute already meant.

That is a faster route and a much quieter one. A bill has a number, a sponsor and a dozen trackers watching it. A board interpretive statement has a PDF in a meeting packet. No floor vote to lose, no session deadline to miss.

The instruction for anyone monitoring health AI: watching state legislatures is no longer enough. Boards move on their own calendar, and their output is what actually reaches a practitioner. Board packets are public. Almost nobody reads them.

What would the AI patient-care policy require of a licensee?

The second draft is the operational one. Its central statement is that licensed physicians, physician assistants and certified anesthesiologist assistants "remain fully and solely responsible for all clinical judgments and patient care outcomes, regardless of whether augmented intelligence (AI) tools were consulted during the episode of care," and that using AI for assessment, diagnosis or decision support "does not diminish, transfer, or otherwise alter a licensee's professional, ethical, and legal accountability."

It also states a baseline fact: "No AI system currently approved or available for clinical use in Washington State has been granted authority to independently assess, diagnose, treat, or prescribe."

Beyond the accountability rule, the draft sets out a human-in-the-loop duty, an affirmative right for a licensee to refuse AI use mandated by an employer or payer, patient disclosure, and attention to training-data bias and skill erosion. Trainees and their supervisors are in scope alongside licensees.

The exclusions are as informative as the inclusions. Administrative automation, backend analytics, threshold alerting, ambient documentation, EMR and billing software, prior authorization and utilization review, and evidence-synthesis tools are all expressly out of scope. The target is narrow: AI that takes part in clinical reasoning. A scribe is not the subject. A differential-diagnosis engine is.

What would the enforcement hook actually be?

The draft states that failure to adhere to its expectations "may constitute unprofessional conduct under RCW 18.130.180 and may subject a licensee to disciplinary action by the Commission." It then closes the other door: "This policy does not create a private right of action and does not establish a civil standard of care for litigation purposes."

Read together, the exposure is specific. No fine, and no theory handed to plaintiffs. It is license discipline, which for a physician is the heavier of the two. A malpractice claim is insured, defended and usually settled. A disciplinary finding is reportable and follows the practitioner across state lines.

The mechanism is familiar. Boards have long treated signing off on work you did not independently evaluate as a conduct problem in its own right. Aimed at a licensee who countersigns AI-generated assessments, it is old law pointed at a new tool.

How are states reaching the same answer by different vehicles?

The interesting variable is not whether AI can practice medicine. Everyone is landing on no. It is which instrument does the work, because that determines who has to be watching.

The split runs along one line. Statutes point at the machine and the company behind it. Boards point at the person holding the license. The legislative track can shut a product out of a state. The board track reaches the clinician, and it moves faster.

Who should act on this, and how?

For a Washington licensee: assume nothing changes today and that something may change on August 21. The useful preparation is knowing which tools in your workflow generate an assessment, diagnosis or treatment recommendation that you then endorse, as against tools that transcribe or bill. Only the first category is in play.

For a health system: the right-to-refuse provision touches policy documents rather than clinical practice. A licensee right to decline an employer-mandated or payer-mandated AI tool sits awkwardly next to a mandate that it be used. Answer that before adoption.

For a vendor selling into Washington: the interpretive statement forecloses autonomous-clinician positioning pre-emptively, before any such product has sought authority in the state. Framing AI as augmenting a named, accountable clinician stays inside the lines. Framing it as substituting for one does not.

One caution on the comment window. The packet states written comments were due "by 5 pm on July 2, 2026," which precedes the July 22 meeting, and the agenda carries a June 08, 2026 date. Quoted as printed; this piece does not reconcile them. Contact the commission directly before relying on either date.

Last verified: July 28, 2026