Health Install #038: The Device Change Log

On Aug. 18, 2026, FDA opened a public discussion on how generative-AI medical devices should be judged before they reach patients — and watched after they are in use.

The useful question is not, “What should I worry about?” It is, “What small household decision gets easier if I make it before the pressure arrives?”

Print this one for your Household Resilience binder. Today’s install is THE DEVICE CHANGE LOG. It takes about 15 minutes and gives one current headline a clear household action.

FDA’s new discussion paper asks for feedback on risk, premarket testing, postmarket monitoring, foundation models, and agentic AI. The agency says GenAI devices may bring risks that are different from older software-based medical devices.

That does not mean your watch or health app is unsafe. It means the old idea of “I bought this device, so I know what it is” is getting weaker. With connected software, what matters is not only the box you bought. It is also the version running inside it today.

By the early 1970s, medical devices were moving faster than the rules built to watch them. Pacemakers, implants, monitors, and new diagnostic tools could do things that had barely existed a generation earlier. Yet federal law still treated many devices far more lightly than drugs.

Congress finally acted with the Medical Device Amendments of 1976. President Gerald Ford signed them on May 28. The law created three risk classes and formal paths for new devices to reach the market. It also gave FDA stronger tools to require records, manufacturing controls, and reports about serious problems after a device was sold.

The important idea was bigger than paperwork. A medical device was no longer treated like a toaster that left the government’s concern the moment it left the factory. Its risks could depend on how it was built, how it was used, what failed later, and what the maker learned after thousands of people had used it.

That is the narrow parallel to today. GenAI devices are not the same as 1970s pacemakers, and FDA is not saying they should be regulated the same way. But the old lesson still holds: when a health tool can change or reveal new behavior after launch, the record cannot stop at the purchase date. The 1976 system built a trail around the device — what it was, how risky it was, and what happened later. A household can borrow that logic in a much simpler way: note the version, the update date, and anything that changed afterward.

Around the first century, the engineer Heron of Alexandria wrote about machines that could seem almost alive. His automata used falling weights, cords, pulleys, axles, water, and air pressure to make doors open, figures move, and small mechanical scenes unfold without a person touching them once the sequence began.

A visitor saw the front end: a door moved or a figure acted. The real story was hidden behind the wall. A weight pulled a cord. A valve changed water flow. A pulley changed direction. The machine’s behavior came from the system underneath, not from the painted object people were looking at.

That made Heron’s machines surprisingly modern in one small way. If you changed the timing, weight, rope length, water level, or valve, you could change what the same visible machine did. University reconstructions of his automata show how simple hidden parts could create a much more complex result.

A GenAI medical device is vastly more complex, and Heron was not building health technology. The narrow lesson is about the system behind the object. Two devices can look identical while the logic inside them is different. Heron’s audience saw the motion; the engineer had to know the hidden sequence. Today, you may see the same screen and same plastic case after an update even though the software underneath has changed. That is why a tiny household log has value. You do not need to understand the code. You only need enough history to say: “This version changed on this date, and this is what I noticed after.”

Across BOTH examples, the pattern is this: when a tool can change, the record has to change with it.

Treat important health technology like a household system, not a one-time purchase. Keep a tiny change log.

Measured improvement: One health tool now has a before-and-after record instead of relying on memory.

Use the official version screen as your receipt. Open the device or app’s About/Version page, take one screenshot, and save it beside your change log. If something feels different later, you now have a simple before-and-after record instead of a guess.

When a tool can change, the record has to change with it.

— James Williamson

Build the small margin before the big pressure arrives.

P.S. Which health tool do you use most: a watch, blood-pressure cuff, glucose device, sleep app, or something else? Hit reply and tell me. Forward this to someone who uses connected health tools.

Sources reviewed for this issue: FDA, Aug. 18, 2026, “FDA Seeks Public Feedback to Inform Regulatory Approach for Generative AI-Enabled Medical Devices”; FDA, “A History of Medical Device Regulation & Oversight in the United States”; RD Milns Antiquities Museum, University of Queensland, “Wondrous Machines: Hero of Alexandria’s Ancient Automata.”