Issue 016 was about what the FDA cleared a machine to say to a patient. This issue is about who is actually writing the rules for AI in health practice, because in a single two-week stretch all three contenders showed their hand. A private coalition shipped a working national program with the AI vendors donating the seats. A senator proposed making the human override of clinical AI a protected right. And HHS, holding more than 7,000 comments asking it to govern, answered with priorities rather than rules. Three actors, three speeds, and not one of them building for a practice of one. That last part is the point of this issue.
Signal 1
CHAI Launched PULSE, a National Program to Put Governed AI Into Public Health Agencies, With OpenAI and Anthropic Donating the Licenses (July 16, 2026)
On July 16, 2026, the Coalition for Health AI, the clinician-founded nonprofit whose governance playbooks underpin the Joint Commission's AI certification, launched PULSE, the Public health Use case and Learning Scaling Engine. The initiative will place up to 2,000 public health practitioners from state, tribal, local, and territorial agencies into structured generative AI pilots, using 10 enterprise licenses donated by OpenAI and Anthropic, with Accenture managing onboarding and playbooks developed across five use cases. A 17-member leadership council selects the participating agencies, and applications run through August 7. This lands on top of CHAI's open-source library, the Applied Model Card and the governance playbooks, which the coalition publishes free on GitHub. CHAI now counts more than 3,000 member organizations. Source: Coalition for Health AI, July 16, 2026.
What this means for you
Watch what just happened structurally. The organization writing the healthcare AI standards is now also running the deployments, selecting the cohorts, and receiving donated licenses from the model vendors themselves. Standard-setting and adoption have become the same motion, at private-sector speed, and every piece of it is designed for institutions: agencies, systems, councils, cohorts. None of it is sized for you, and that is not a criticism, it is a reading. CHAI holds the standards seat. The practitioner's job is not to join a cohort. It is to know what the standard requires and translate it down to a practice of one, because when those playbooks harden into expectations, they will not come with a solo edition.
Signal 2
Senator Markey Released an AI Accountability Agenda Whose Healthcare Plank Would Make the Human Override of Clinical AI a Protected Right (July 10, 2026)
On July 10, 2026, Senator Edward Markey of Massachusetts released The AI Accountability Agenda, a package of roughly a dozen bills addressing AI risks across sectors. The healthcare plank, titled Putting Humans First in Healthcare, calls on Congress to pass the Right to Override Act. The bill would require healthcare facilities and health plans to create a human override option for AI decisions, and it would protect healthcare workers who disagree with an AI recommendation from retaliation when they act on their own professional judgment for the patient. The agenda arrives as a proposal, not a statute. What it documents is the assumption behind it: that institutions deploying clinical AI may pressure the humans inside them to defer to the machine. Source: Office of Senator Edward Markey, press release, July 10, 2026.
What this means for you
Every prior human-review rule this Brief has tracked told institutions to insert a human into the loop. This one is different in kind. It treats the clinician's disagreement with the AI as something that needs legal protection, the way whistleblowing does. Notice what that concedes about deployed clinical AI: institutional pressure runs toward accepting the output, because accepting is fast and overriding creates friction. Here is the part that matters for a solo practice. You are the institution. Nobody can retaliate against you for overruling a tool, which means the one protection Congress thinks employed clinicians need is one you already hold. The question is whether you use it, or whether the convenience of accepting the output governs you instead.
Signal 3
More Than 7,000 Comments Asked HHS How to Govern Clinical AI, and HHS Answered With Priorities, Not Rules (Reported July 1, 2026)
In December 2025, HHS issued a request for information on using its regulatory, reimbursement, and research authority to accelerate AI in clinical care. By the late-February deadline it had received more than 7,000 comments from providers, researchers, and industry groups. In a briefing reported July 1, 2026, HHS officials summarized what the sector asked for: coordination across HHS agencies, support in implementation and in creating governance structures, and guidance on what makes a good AI tool and which tools actually work. HHS deputy chief AI officer Arman Sharma framed the agency's response around building trust before adoption, and pointed to early steps rather than rules: research programs, developer competitions, and FDA work on future policy for autonomous AI tools. Source: Healthcare Dive, July 2026; HHS RFI, December 2025.
What this means for you
Read what the commenters asked for one more time. Governance help. Evaluation benchmarks. Guidance on which tools work. These are hospital systems with compliance officers, legal teams, and IT departments, and their message to the federal government was that they do not know how to govern this technology on their own. If that is where the best-resourced buyers in healthcare stand, the vendor telling you a tool is safe because everyone is adopting it is describing a market that just asked for help, not a market that has answers. The guidance those 7,000 comments requested will arrive on government time. Your patients are using AI on this afternoon's time.
The Pattern
Line the three signals up by speed. The private coalition is already deploying: programs, cohorts, playbooks, donated licenses from the model vendors, all in motion this month. Congress is proposing. The agency is summarizing comments. The standards seat in healthcare AI is being occupied at private-sector speed while the public rules move at government speed, and everything shipping from that seat is built for institutions. PULSE is for agencies. The Override Act is for facilities and health plans. The HHS guidance, when it comes, will be for the sector. Nobody in this chain is writing the solo edition, which means governance at the scale of one is translation work, and the translation is yours. That is not a burden the big systems escaped. It is the one advantage a small practice holds: you can read what the standard requires and be running it by Friday, while the institutions are still scheduling the committee that decides who attends the meeting.
One Thing You Can Do This Week
Before the next AI tool pitch reaches you, look up CHAI's open-source Applied Model Card. It is a short, free, public disclosure format covering what a model was trained on, tested for, and where it fails. Then ask your current or prospective vendor for theirs. The emerging standard is public, so a vendor who cannot produce one is telling you exactly where they stand relative to it, and that answer costs you one email to obtain. Keep the reply with your practice records. That is vendor governance at the scale of one, done this week.