An 89-year-old in a narrow old neighbourhood in Wuhan presses a call button on her wristband, and a community worker reaches her soon after.
In a different community nearby, an elderly man living alone falls at home. His emergency bracelet triggers. A community care adviser is at the door in five minutes.
Behind both responses sits a stack of hardware no marketing brochure mentions: door sensors that notice when a front door has not been opened for days, water sensors that watch for a forgotten tap, smoke alarms, and a thin mat under the bed sheet that reads heart rate and pulse from above.
Key takeaways
- Wuhan's Wuchang district has placed more than 750 AI devices inside its care homes, on top of 1,239 home care beds fitted with smart devices, according to district and civil affairs reporting.
- The same district reports handling close to 80,000 smart emergency calls and roughly 3,000 in-person follow-up visits, on a platform built for about ¥3 million (≈ US$420K / HK$3.3M).
- The city has committed to fitting emergency call and monitoring devices in the homes of 11,000 willing isolated elderly residents in 2026, approved by the municipal people's congress.
- Xinhua's reporting describes the operating principle plainly: devices handle standardised monitoring, while washing, skin checks, emotional support and home visits still require nursing staff face to face.
- The unresolved question is consent. Some pilot communities have dropped conventional cameras in favour of vision-sensing devices that draw a human outline rather than a face, and residents get a physical switch to turn monitoring off.
What eldercare AI actually looks like
This is the least glamorous corner of the AI economy, and possibly the most demanding. The population is old, often frail, sometimes cognitively impaired, and rarely a willing user of a new device. China's bet is that eldercare (养老) can be made safer by instruments rather than by more hands, and that the instruments are the cheaper half.
Wuhan's approach starts with the building rather than the person. In Wuchang district, a district-level eldercare platform ties together city, district, street and community service networks, with three-tier accounts for administrators, providers and residents, so a service order can be placed, scanned on arrival, video-verified, quality-checked and settled without a paper form.
Inside the institutions, the equipment list is unglamorous: a physiotherapy bed that builds an individual recovery plan after basic health information is entered, a mechanical-arm moxibustion device, a smart pillow that adjusts support when it detects snoring, and a small robot that patrols a route and sends photographs to a phone.
The district also runs what it calls an artificial-intelligence eldercare experiment, with six experimental communities and eight "silver" community centres, and a social experiment with 100 households across five communities tracked over five years.
The numbers that matter
The figures Wuchang publishes measure response rather than novelty.
- More than 750 AI devices deployed inside care homes since the pilot began.
- 1,239 home care beds built, each with one-touch call devices and sensors for activity and vital signs.
- Close to 80,000 smart emergency calls handled, and about 3,000 door visits made in response.
- A four-level telephone response chain, with a 24-hour hotline at district level.
A district care platform costing roughly ¥3 million (≈ US$420K / HK$3.3M) also funds roughly 200 elderly care advisers and links more than 100 service companies, and runs the "one-click call" system with alerts to emergency contacts.
At city level the scale is bigger. Wuhan's unified eldercare platform has gathered more than 2,000 service providers and around 14,000 products, and the municipal government's 2026 public-welfare list, approved by the people's congress, commits to installing emergency call and monitoring equipment for 11,000 willing isolated elderly households, plus 120 new community facilities.
Where the sensors stop
Xinhua's coverage of Wuhan carries the most useful line in this story, from the head of a local eldercare company: at this stage, eldercare AI is not machines simply replacing people, but artificial intelligence and human care working together. The devices take on standardised needs such as home safety monitoring and basic health warnings; washing, dressing, skin checks, emotional counselling and home companionship still need a caregiver in the room.
That boundary is the honest one. A wristband that knows you have not opened your fridge cannot tell you whether you are lonely. It can only call someone who can.
According to Li An, chief scientist at BrainNet (脑机网), the meaningful frontier in eldercare AI is not a smarter sensor but a closed loop: a signal detected, a person reached, and an outcome somebody checks.
The privacy question is being fought in practice rather than in policy papers. Local reporting describes pilot communities abandoning ordinary surveillance cameras in favour of devices that sense the shape of a person rather than recording a face, and staff pointing to a switch and telling the resident: this switch is yours, turn it off when you do not want to be watched.
Honest limitations
The counts here come from local government, district civil affairs reporting and Xinhua coverage, none of it independently audited, and several figures are cumulative since launch with no denominator for the population being served. Close to 80,000 emergency calls sounds impressive until you know how many elderly residents are actually enrolled in the scheme in Wuchang, which the reporting does not disclose.
The 750 devices in care homes is a deployment figure, not an adoption figure. Nobody has published how often they are used, whether staff find them reliable, or how many false alarms they generate. The 5-minute response to a fall in one neighbourhood is an anecdote, repeated precisely because it works.
The city-wide platform figures deserve particular caution. One set of official answers cites 1.54 million registered users and more than 2 million service orders for Wuhan's eldercare platform; another official document, published a month later, gives 1.277 million registered users and 1.763 million service visits. Both come from city agencies, the figures do not match, and the later-dated one is lower on both counts.
The deeper limitation is that this category measures activity, not outcomes. A call answered in five minutes is a process metric. Whether the elderly person was safer or less likely to end up in hospital is not something any of these deployments has published.
What readers can do now
- If you are setting up monitoring for an elderly relative, ask whether the device records images or only senses presence, and insist on a physical switch. Local pilots in Wuhan have started giving residents exactly that switch, and a system nobody can switch off is surveillance, not care.
- Test the response chain before you need it. A wristband that triggers a phone call is only useful if someone answers it within minutes at three in the morning; ask the provider what happens when the answer does not come.
- Treat any institution's "AI elderly care" claim as a claim about monitoring hardware unless it can name what happens when the monitor fires. The part of eldercare needing hands, patience and a familiar face is the part no sensor replaces.