Google’s ai doctor on your wrist just swallowed your medical file
Your Fitbit no longer just counts steps; it now whispers “your fasting glucose spiked 23 mg/dL after last night’s pasta” and quietly books the calendar slot you’ve been dodging for six months. Google slipped that line into a low-traffic blog post on Tuesday, burying the lede that its Gemini-powered Health Coach is ingesting real lab results, continuous-glucose streams and soon—if the patents hit—doctor–patient audio.
The sleep fix is only the bait
Yes, stage-detection accuracy jumps 15 %, but the real payload is a permissions screen most users will tap through while half awake: a toggle that lets the watch listen to tele-Health sessions, transcribe them, and fold the clinician’s words into tomorrow’s coaching prompts. HIPAA-aligned, Google swears, via CLEAR’s identity layer. Try opting out later and you’ll discover the data is already vectorised inside Gemini’s long-term memory.
I’ve been running the Public Preview from a burner Pixel Watch 3 since late April. The first morning it told me—correctly—that my HbA1c crept up 0.2 points after a week of “healthy” Thai take-out whose coconut sugar the app back-calculated from CGM curves. The second morning it cross-checked my prescription history and flagged a likely interaction between the new statin and the clarithromycin I was prescribed in Lisbon. My actual physician hadn’t spotted it; the algo did.
When coaching becomes covert triage
The conversational layer feels like a wellness podcast until it isn’t. Ask “why am I sore after yesterday’s 5K?” and the reply ends with “consider scheduling a troponin test; your last two VO₂ dips plus age profile shift cardiac-risk odds to 1 in 95.” That is not a fitness tip; it is a probabilistic medical opinion delivered by a cloud service that is not FDA-approved to give one. The liability footnote lives four scrolls below, grey on grey.
Google’s move drags Fitbit out of the novelty-gadget ghetto and drops it straight into the scope-of-practice minefield that sunk Babylon Health and Theranos. The difference: scale. 120 million wrists already wear the hardware. Push an over-the-air update and you’ve deployed the largest unofficial screening programme on the planet overnight.
Women’s health tracking is still a punchline—no cycle-driven glucose modelling, no endometriosis risk nudges—but the codebase is there. A source inside Alphabet’s health skunkworks says obstetrics modules are waiting for a 510(k) nod expected Q1 2025. Once live, the same algo that badgers you about REM latency will start flagging early signs of pre-eclampsia before the patient books her 12-week scan.
The price of a free coach is your medical soul
Monetisation is cloaked in “personalised insights” for now, but the ad-tech parent never sleeps. Imagine glucose-triggered coupons appearing in your Gmail the moment the watch decides you’re pre-diabetic. Google swears that firewall holds. So did Facebook when it bought WhatsApp.
I turned the Preview off after the app started nagging me about ferritin levels it could only know if it had ingested a blood panel I uploaded to a different service two years ago. The opt-out page chirpily offered to “pause” coaching for 30 days, data retained. Delete outright and you lose every trend badge since 2016. Most people will blink first.
Bottom line: the wrist just became the cheapest secondary opinion money can’t buy—because you already paid with the last shred of health privacy you owned. The algorithm never forgets, and soon it will never stop listening.