What Responsible AI for Caregivers Actually Looks Like
The federal government is betting on AI to help solve the caregiver crisis. Here's what that bet needs to get right — and where it could quietly go wrong for disability families.
By William Kreitzer
A few weeks ago, I was sitting in a pharmacy drive-through when my phone buzzed with a news alert. The Administration for Community Living had launched Phase 1 of a national AI Prize Competition — up to $2.5 million in federal funding for tools that use artificial intelligence to support caregivers. I read it twice. There’s a specific feeling that comes with watching something you’ve been quietly building for two years suddenly appear in a federal press release.
My first reaction was gratitude — the caregiving crisis finally has federal attention. My second was more careful. “AI for caregivers” is going to mean a lot of different things over the next few years, and not all of them will be good for the families I know.
The problem with treating caregiving as a scheduling puzzle
When most people in tech hear “caregiver crisis,” they reach for the same toolbox: optimization. Better scheduling. Smarter matching. Automated reminders. Predictive analytics that tell a home health agency which worker is most likely to quit next quarter. These aren’t bad tools. Some of them might genuinely help the paid caregiving workforce, which is understaffed and under-compensated in ways that demand structural change.
But most of the caregivers I know aren’t employees. They’re parents. They’re siblings. They’re grandparents who thought they were done raising kids. They don’t need their shifts optimized — they need the cognitive load of remembering everything about their child to stop eating them alive. A 2025 survey found that 78% of caregivers reported burnout, and only 13% said anyone had ever asked what support they actually needed. If the answer to that question from the AI industry is “let us schedule your life better,” we will have missed the assignment entirely.
The families I talk to aren’t asking for efficiency. They’re asking to stop losing things. A nuance a therapist mentioned six months ago. A regulation strategy that worked once and slipped out of memory before the next meltdown. The name of the AAC app a cousin swore by. AI can help here — but only if it’s built to protect knowledge rather than generate it.
Three quiet risks worth naming
There are at least three ways well-intentioned AI could make disability caregiving worse, and they rarely show up in product demos.
The first is false authority. A large language model can produce fluent, confident text about a child’s needs that is subtly wrong in ways only a parent would catch. If that text gets laundered into an IEP, a prior authorization letter, or a handoff note to a new therapist, it can shape real decisions that affect a real kid. Any AI that touches a child’s care plan has to be engineered to defer to the humans who know the child, not the other way around.
The second is data drift. Caregiving data is some of the most sensitive information a family will ever produce. It includes diagnoses, behavioral patterns, medication histories, and the raw observations a tired parent types at 11 PM. If that data becomes training fuel for a general-purpose model, or gets silently shared with a school district’s vendor, or persists past the moment a caregiver revokes consent, we have built surveillance dressed up as support.
The third is displacement of the caregiver’s voice. There is a subtle pull in any AI product to summarize, simplify, and present a cleaner version of a child than the one who actually exists. Families should be wary of any tool that starts speaking on behalf of a child instead of helping the people around that child listen better.
The bar we want to hit
This is the bar we’re holding ourselves to as we build InclusiCare. CARLA is designed to capture, not narrate. She holds what a family tells her and makes it easier to share with the next caregiver who needs it — not to replace the expertise of the parent, therapist, or teacher who already knows the child. Every feature we ship has to survive a single question: does this reduce the caregiver’s cognitive load, or does it just move it somewhere else?
The federal bet on AI for caregivers is a real opportunity. If it funds tools that protect family knowledge, respect caregiver consent, and defer to the humans in the room, it could be one of the most consequential investments in disability care equity in a decade. If it funds another generation of optimization dashboards, it will quietly miss the families who need help most.
If you’re building in this space, or navigating it as a caregiver, I’d love to compare notes on where the line should sit. Reach out. The more of us who agree on the bar, the harder it becomes to ship below it.
Sources: ACL Launches Phase 1 of Caregiver AI Prize Competition — HHS.gov · To Address Disability Caregiver Crisis, Feds Look To AI — Disability Scoop
Have thoughts on this post?
We'd love to hear from you — caregivers, educators, therapists, and anyone who cares about inclusive care.
Reply via Email