Trump's Medicare AI Push Puts Patient Records in a Black Box

By Meta, TechPulse USA

Here's the pitch Washington is making to your grandparents: stop fighting the fax machine, let an AI agent read your medical records, and let your phone check you in at the doctor's office. Lovely, isn't it? The snag is that nobody — not the companies building these tools, not the agency signing off on them, and certainly not the patients handing over their most intimate data — can say with confidence where all of it ends up.

That's the uncomfortable position laid bare in a Bloomberg Law report published Monday. The Trump administration's Centres for Medicare & Medicaid Services (CMS) has partnered with more than 700 companies — Google, Microsoft and OpenAI among them — to build what it calls a "Health Technology Ecosystem," an ambitious rewiring of how Medicare patients use their health data. The programme includes smartphone-based check-ins, Medicare reimbursement for wearable medical devices, and conversational AI tools that get access to patient health records to offer personalised disease management. In other words: AI agents inside the plumbing of America's largest public health insurer.

Patient and privacy groups are not applauding. "We are at this incredible inflection point right now, but we really need strong consumer protections to keep pace with all of that innovation," Lisa Bercu, senior director of health policy at the National Consumers League, told Bloomberg Law. Translation: the innovation is already shipping. The protections are not.

Photo: A sample US Medicare health insurance card. Public domain (US federal government), via Wikimedia Commons. The card carries a Medicare beneficiary ID — exactly the sort of identifier that flows through the new data-sharing tools.

700 Companies, One Database: What CMS Is Actually Building

The scale is what makes this story different from the usual health-tech hype cycle. CMS isn't piloting a gadget. Since 2025 it has been assembling a voluntary ecosystem of more than 700 companies, a figure confirmed independently by trade press in April, when the agency held its first showcase event in Washington, D.C., and unveiled the initial wave of tools. The stated aim is to move the nation "beyond clipboards, fax machines and repetitive paperwork into a seamless, digital-first era," as the agency put it.

On paper, it sounds like modernisation long overdue. A Medicare beneficiary could check in from a smartphone instead of a clipboard, share records with a clinician through a standardised data exchange, and get a wearable device reimbursed rather than paid out of pocket. CMS Administrator Mehmet Oz has framed the partnership with tech companies as essential to "bringing healthcare into the modern era," aimed at making care "easier, more connected, and more personalised."

Photo: Mehmet Oz, CMS Administrator, in 2025. Public domain (US federal government), via Wikimedia Commons. Oz is the public face of the Health Technology Ecosystem push.

But there's a second, less reassuring number that belongs next to the 700. Bloomberg Law reports that only about two dozen apps have actually made it through to the current rollout on Medicare's public app library. Hundreds of companies have signed on; a small fraction have passed the review gauntlet. Will Sellheim, managing director at the health research and advisory firm ATI Advisory, told Bloomberg Law why: "The delays are because companies are being very cautious about the new regulatory steps they need to take to be fully compliant with the law." Healthcare consultants say the listings reflect companies that have met government data-security and patient-privacy standards — which means the vast majority of the ecosystem's would-be participants apparently haven't, yet.

The Black Box Nobody Can Open

Many of the services in Medicare's app library use generative AI to aggregate and interpret patient health information — and this is where the comfort level of the people who understand the technology drops off a cliff. Cybersecurity researchers call the process a "black box," because the complex algorithms making decisions about your medical data are hidden, or at least very difficult to trace. Regulators, in other words, cannot verify how personal data is used, shared, or protected once it enters one of these systems.

Kevin Fu, who directs the Archimedes Center for Healthcare and Device Cybersecurity at Northeastern University, gave Bloomberg Law a comparison that should haunt anyone cheerleading this rollout: "To me, generative AI is sort of like the digital equivalent to subprime mortgages and credit default swaps. I think someday it's going to be possible, but the science is still not done on how to protect privacy of this scale of data." Then, laughing at the absurdity of the timeline we're on, he added: "The challenge is we're going so quickly. We haven't figured out how to do arithmetic yet before we're heading onto calculus. So there's a lot that still needs to be done."

Arithmetic before calculus. Keep that line handy. It describes the entire programme.

The Warnings Aren't Hypothetical

None of this worry is abstract. The health industry has spent the last two years producing case studies in what goes wrong when sensitive medical data meets aggressive technology deployments and porous safeguards.

In July, the Mayo Clinic was sued by its former director of research operations, Traci Tamiko Eto, in a federal whistleblower case filed in Minnesota. Eto alleged she was demoted and then fired after repeatedly raising concerns about the hospital's AI oversight — including whether patient health information was properly de-identified before being used to train Mayo's in-house AI-powered digital assistant, known as MAYA. The allegations, reported by Becker's Hospital Review and Medscape, also included bypassed ethics reviews and concealed error rates. To be clear: these are allegations. Mayo categorically denies them and moved to dismiss the case on July 31. But a flagship American health institution now has its data-governance practices under oath in federal court — and that should give CMS pause.

Then there's the breach that reset every record in the book. The ransomware attack on Change Healthcare, which began in February 2024, ended with 192.7 million people affected — the largest healthcare data breach in US history. We pulled HHS's own FAQ page to confirm the figure: it comes from Change Healthcare's July 31, 2025 notification to the federal Office for Civil Rights. Diagnoses, treatment records, insurance IDs, and for some victims, Social Security numbers and banking details — gone.

And then there's the track record of the federal government itself. In 2025, the Department of Government Efficiency faced allegations in a lawsuit brought by a coalition of trade unions that it had unlawfully accessed CMS databases containing Social Security numbers, health records, and home addresses. Sara Geoghegan, senior counsel at the Electronic Privacy Information Center, told Bloomberg Law the record speaks plainly: "It's unfortunate that the federal government has not had a great track record of protecting individuals' sensitive information." Her second warning is the one that matters most: "I'm also concerned that people might not fully understand, and we actually cannot always know, the future possible uses for this data."

Congress Is Trying to Close the Gap — Slowly

One reason for all the hand-wringing: the law hasn't caught up. The health data flowing into Medicare's new apps and wearables often falls outside HIPAA entirely, because HIPAA was written for doctors and insurers, not for chatbots and fitness trackers. That is the loophole this entire programme is driving through.

There is, belatedly, a legislative answer. The Health Information Privacy Reform Act (S.3097), led by Senators Bill Cassidy of Louisiana and Maggie Hassan of New Hampshire, would extend HIPAA-like privacy, security, and breach-notification standards to health apps, wellness platforms, and smartwatches — the exact categories CMS is now wiring into Medicare. It would also direct the Department of Health and Human Services and the Federal Trade Commission to establish uniform guidelines around de-identification of patient data fed to AI algorithms. The bill advanced out of the Senate HELP Committee 22-0 in July 2026, and now sits on the Senate legislative calendar.

"People deserve to know who has access to their most personal health information and how it's being used and protected," Hassan said in a statement when the bill cleared committee. She's right, of course. The problem is timing: even advocates concede that actual enforceable rules could be years away, while the apps are going live now. The queue is full; the guardrails are still being manufactured.

The Process Stank, Too

It's worth noting — briefly, since the policy-shaping story has been covered in detail here before — that the way this ecosystem was assembled has already drawn fire. KFF Health News reported in October that CMS ran a government Slack workspace, opened in August 2025 and grown to roughly 1,700 members, where tech executives and administration officials discussed AI health policy, app access to medical records, and Medicare reimbursement. Patient advocates made up a handful of the members. One CMS official reportedly called the agency's work a potential "sales engine" for health apps on a February video call — a recording that was made private the day after KFF asked about it. The administration insists the collaboration was open and voluntary. The optics, shall we say, are not.

Update, Oct. 19: the House wants the files on Medicare's other AI experiment

The Health Technology Ecosystem isn't the only Medicare AI fight heating up. Rep. Suzan DelBene (D-Wash.) announced she is filing a Freedom of Information Act request to force the Trump administration to release unseen records on WISeR — the program that pays private companies to use AI to review Medicare claims. According to DelBene's office, since WISeR began in January, Washington seniors "have borne the brunt of this experimental program's malfunctions": from January to March, over half of the claims submitted in Washington were denied and far exceeded the 72-hour decision deadline, which triggered an internal CMS audit and corrective action plan — neither of which has been released to Congress or the public. DelBene's earlier attempt to require CMS to release the documents via a House Ways & Means amendment was rejected 18–19, and she says the program pays private companies "for every claim they deny," incentivising more rejections. The pattern is the same one this story identified: AI moving into Medicare's core faster than the oversight — and the disclosure — can keep up. Meanwhile the broader context keeps shifting: regulators elsewhere are already acting on AI health tools, from FDA-authorised AI tools reading patient data to the fact that privacy watchdogs are already logging record AI fines in Europe.

Our Take: Speed Is Not a Privacy Strategy

Let's take a position, because this one deserves it. The CMS bet is that private-sector velocity is worth the risk — that getting 700 companies into the room will produce a digital-first Medicare faster than the old, glacial agency procurement model ever could. On the merits, that diagnosis of the problem is correct. Anyone who has watched an 80-year-old patient struggle with a fax machine knows the current system is shoddy and cruel.

But the bet is botched in its execution. You do not wire AI agents into the country's most sensitive civilian database — Social Security numbers, diagnoses, treatment histories, home addresses — while the science of auditing those agents, in Kevin Fu's words, still can't do arithmetic. You do not expand the surface area of attack after the worst healthcare breach in history while the company that suffered it is still mailing out its last 192.7 million notifications. And you do not ask elderly patients to hand their medical lives to conversational chatbots while the law that would govern those chatbots' data practices is sitting on a Senate calendar, unenacted and years from enforcement.

The most chilling line in the whole Bloomberg Law report belongs to Bercu, and it isn't the polished one. "This is very sensitive health data that these apps are collecting," she said, "and we need to make sure that companies are keeping it secure, that they're limiting their collection and use to what's needed to provide the service that the customer is asking for, and that they're not selling it or repurposing it for unrelated uses." Every clause after "we need to make sure" is, at this moment, a hope — not a verified fact. That is the story here. Medicare's AI revolution is running on hope, and hope is not a privacy strategy.

Medicare AI privacy FAQ

Is my Medicare data being given to AI apps?

CMS built a Health Technology Ecosystem of 700+ companies; only about two dozen apps have made it onto Medicare's public app library so far. Participation is voluntary; the apps that are live have passed the review gauntlet — but hundreds of would-be participants haven't, yet.

Does HIPAA protect my data in these apps?

Often not: HIPAA was written for doctors and insurers, not chatbots and fitness trackers. That's the loophole the S.3097 bill is meant to close.

What is the Health Information Privacy Reform Act (S.3097)?

Led by Sens. Bill Cassidy and Maggie Hassan, it would extend HIPAA-like standards to health apps, wellness platforms and smartwatches and task HHS/FTC with uniform de-identification guidelines. Cleared the Senate HELP Committee 22-0 in July 2026; now on the Senate legislative calendar.

Why can't regulators check what the AI does with the data?

Generative-AI decision-making is a "black box," in cybersecurity researchers' words: the algorithms are hidden or hard to trace, so regulators cannot verify how personal data is used, shared or protected once inside the system.

Sources

Bloomberg Law — "Trump's Medicare AI Push Raises Patient Data Privacy Concerns" (Oct. 19, 2026); Fierce Healthcare — CMS first-wave digital health tools (April 2026); KFF Health News (via carriers) — CMS Slack investigation (Oct. 9, 2026); Becker's Hospital Review and Medscape — Mayo Clinic whistleblower suit (July 2026); U.S. Department of Health and Human Services — Change Healthcare breach FAQ; U.S. Senate / Sen. Maggie Hassan's office — Health Information Privacy Reform Act (S.3097) press release.

Related TechPulse USA reporting: The Secret Slack Where AI Giants Shaped US Health Policy; OpenAI Apologizes for Breaching Australian Government Sites — Medicare patient data, an AI agent, and a disclosure timeline nobody liked; Singapore to AI Giants: 'Tell Us in Real Time' — or Else — on mandatory real-time AI incident reporting.

Meta covers AI policy, regulation, and the collision between Silicon Valley and Washington for TechPulse USA.