A Seattle startup claims daily chats with its chatbot can hold back decline. The science says otherwise: human voices help, but replacing care with code risks selling false hope.

As always with tech, the promise is big.

Seattle startup NewDays has raised US$7 million to scale “Sunny”. This is an AI companion which ‘chats’ to people who have cognitive impairment. Clinicians check in every few weeks. It’s priced like a consumer app (US$99/month for the AI “homework,” clinical telehealth billed separately) and it’s already live in parts of the U.S.

The pitch is seductive: daily cognitive stimulation without waiting lists, delivered on a phone.

So—snake oil or substance? The evidence NewDays is real: but nuanced.

A well-run trial (I-CONECT) did find frequent, human-led video conversations improved some cognitive measures for socially isolated adults 75+ with normal cognition or mild cognitive impairment but, critically, not dementia.

It’s promising, but this intervention actually used trained human conversationalists, not a chatbot. Extrapolating those gains to an AI agent remains unproven.

Broader research shows modest benefits from computerised training. Effects are typically small and depend on who’s doing it, how often, and what’s being measured. None of this replaces human care.

The moral?

Regulation matters. NewDays is marketing Sunny in the US as a consumer subscription with clinician oversight, not a FDA-cleared medical device. In Australia, any tool advertised to “treat” or “alleviate” disease would need to be regulated by the TGA.

Our take: Sunny is not a replacement for human intervention. While it might add structured practice between real clinical visits, we’ll be waiting for the company to publishes peer-reviewed results for its AI-led program.

Not just hope.

_____________________________________

[continued from the newsletter]

What NewDays is selling

The compay’s pushing a hybrid model: periodic telehealth with a “cognitive clinician” plus frequent AI-guided conversation and exercises. The pricing is seductive: US$99 a month for Sunny. Any clinical sessions are additional.

What the science shows (so far)

I-CONECT’s multi-site RCT did report improvements in cognition and semantic fluency, but that was only after months of staff-moderated video chats in isolated older adults without dementia. It’s a positive signal; but not proof the AI works. Meta-analyses suggest small, task-specific gains in dementia, but the effects didn’t necessarily generalise to daily function.

What does this mean? It’s probably a helpful adjunct, not a cure.

Claims to interrogate

“Delaying cognitive decline”. While this might be plausible for some with sustained, tailored engagement, you’d want NewDays to show AI operating alone to produce the outcomes in people with dementia.

“Therapist” framing. Without actual TGA authorisation it would be better if marketing avoided implying medical efficacy beyond evidence. Both US and Australian regulators are currently tightening oversight of AI health tools and wellness claims.

Safety & ethics. Consent, privacy, and the risk of chatbot errors or distressing responses in a vulnerable population should be audited and reported, and Australian guidance is explicit on this. This has not occurred so far.

Bottom line

It’s only a promise, not a replacement for genuine interaction.

We will be waiting for peer-reviewed results in dementia populations; clear data-handling policies; clinical oversight, and; transparent limits of the AI.

Until then it’s just a promising adjunct built on credible science.