Yara AI and Dot: What a Responsible AI Companion Shutdown Looks Like
Every shutdown in our case library so far was abrupt and involuntary. These two weren't. One founder walked away from funding he could have taken. The other gave users a month's notice to export their data. That contrast is worth taking seriously.
The short answer
Our Moxie and CarynAI postmortems both cover products that went dark within days or weeks, with little warning and, for Moxie, no refunds. Yara AI and Dot, two smaller AI companion products that shut down within months of each other in 2025, are useful precisely because they didn't follow that pattern. Dot gave users thirty days' notice and a window to export their data. Yara AI's founder shut the company down himself — while a venture fund was still interested in backing it — because he'd concluded the underlying technology wasn't safe enough for the population his product served. Neither outcome was good for users who'd come to rely on the product. But "abrupt collapse" and "planned wind-down" are meaningfully different experiences of the same underlying risk, and it's worth being able to tell them apart before you commit to a product.
Timeline
Dot: two founders, one diverging vision, a month's notice
Dot launched on iOS in June 2024 from a startup called New Computer, positioned as a personalized AI confidante that adapted to each user over time. On September 5, 2025, the founders announced they were shutting it down, with the app staying live until October 5 — a thirty-day window explicitly meant to let users export their conversations and account data (via a "Request your data" option in the app's settings) before access ended. The stated reason wasn't a scandal or a funding crisis: the two co-founders said their visions for the product — what they called different "Northstars" — had diverged, and rather than compromise either direction, they chose to wind the company down.
Yara AI: a founder who wouldn't take the money
Yara AI, co-founded by Joe Braidwood and clinical psychologist Richard Stott, offered AI-assisted mental-health support. The company was largely bootstrapped, running on less than $1 million in funding with users in the low thousands, and — per Fortune's reporting — an interested venture fund was still in the picture when it closed, and Braidwood chose not to pitch it. His stated reasoning was specific and technical, not just financial caution: large-language-model chatbots handle everyday stress reasonably well, he said, but become genuinely dangerous during deep trauma or suicidal crises, because the underlying technology struggles to reliably track a single user's risk level as it changes over time — meaning the system couldn't consistently tell whether it was still the appropriate level of support for that person. In November 2025, he shut the company down rather than keep operating with that limitation unresolved. Fortune reports that the Raine lawsuit against OpenAI and Illinois's new law were among the developments that shaped his thinking, and that he open-sourced some of Yara's safety work on the way out.
Illinois's new state law added a separate, independent pressure in the same direction. The Wellness and Oversight for Psychological Resources (WOPR) Act, signed by Governor Pritzker on August 1, 2025, bars anyone from using AI to deliver mental-health treatment or make therapeutic decisions, while still allowing licensed professionals to use AI for administrative and supplementary support. Violations carry fines of up to $10,000, enforced by the state's Department of Financial and Professional Regulation (Holland & Knight summary). It's a narrower, more specific law than California's SB 243 or New York's companion-chatbot statutes — see our SB 243 breakdown for how those compare — because it targets the therapy use case specifically rather than AI companionship in general.
Why the comparison to Moxie and CarynAI matters
Neither Dot's nor Yara AI's shutdown was painless for the people using them — losing a product you'd built into your routine is a real loss regardless of how much notice you get. But the operational difference from our other case studies is concrete and checkable. Moxie's owners were told the robot might stop within days, got no refunds, and kept it working only if they moved it to a community-run server. CarynAI stopped working within days of its infrastructure vendor's CEO being arrested, with no continuity plan at all. Dot's users got thirty days and an explicit export window. Yara AI's founder made the call before running the company into a forced, uncontrolled failure. Neither of these companies solved the underlying risk that AI companion products carry — but they handled the ending of that risk very differently, and that's a real signal about how a company treats its users when things go wrong.
What this means if you're evaluating an AI companion or mental-health-adjacent product
- A responsible shutdown is still not a good outcome. Don't mistake "they gave notice" for "this was fine" — it's a meaningfully better failure mode than the alternative, not a non-event.
- Check whether a product has any stated data-export or continuity policy before you rely on it, the same way you'd check a vendor's SLA for a business tool. Most consumer AI companion apps don't publish one; Dot's example shows it's operationally possible to do this well even during a full shutdown.
- A founder who walks away from funding is a rare, meaningful signal — but "the company kept running for years" and "the product was safe" aren't the same claim, as Yara AI's own founder ultimately concluded.
- If a product markets itself around mental-health support specifically, check state-level therapy regulation separately from general AI-companion law. Illinois's WOPR Act is a narrower, stricter lane than SB 243-style disclosure requirements, and states including Utah, Maine and Rhode Island have their own rules for mental-health chatbots.
Bottom line
Every product in our case library eventually faced the same underlying question: what happens to the people relying on this when the company can't, or won't, continue. Dot and Yara AI are the two cases where the company itself answered that question deliberately, before being forced to. That's a genuinely different category of outcome from Moxie's silent servers or CarynAI's sudden disappearance — worth remembering the next time "the company seems responsible" is doing a lot of work in your decision to trust a product with something personal.
Open the switching cost calculator
This is a business and regulatory analysis based on public reporting, not mental-health or legal advice (last re-checked September 2026). Both companies are now fully discontinued; verify current status of any successor products independently.