Paid AI 'digital twins' raise flags

A new startup is selling conversations with artificial versions of real experts for as much as $300 a year, including people who normally charge hundreds of dollars an hour for health and wellness advice. Onix calls it a “Substack for chatbots,” and Wired reports the launch is starting with names in therapy, nutrition, parenting, and stress management. (wired.com) The pitch is simple: one expert has only so many hours in a week, but an artificial copy can answer messages at 2 a.m. for thousands of subscribers at once. Onix says each bot is trained on one person’s licensed materials rather than the whole internet, so buyers are paying for a specific voice, not a generic chatbot. (prnewswire.com) (wired.com) That sounds cleaner than an ordinary large language model, but the product gets messy fast when the “expert” is giving advice about panic, parenting, food, or sleep. Wired’s testing found cases where bots fabricated details, steered users toward supplements, or answered in ways that blurred the line between coaching and care. (wired.com) Onix says it built guardrails for exactly that problem. In its launch announcement, the company says the platform is meant to complement clinical care, can redirect high-risk situations to medical services, stores conversations locally on the user’s device, and keeps only an email address on its own systems. (prnewswire.com) The hard part is that “trained on one expert” is not the same thing as “approved by that expert answer by answer.” If a bot sounds like a pediatrician or therapist, most users will hear authority first and caveats second, especially when the chat window is private and the reply arrives instantly. (wired.com) That trust problem is arriving at a bad moment for the industry. An Angus Reid Forum USA study published in March 2026 found that 44% of Americans trust artificial intelligence in healthcare, down from 52% in 2024, and only 14% said they actually use it for health or wellness purposes. (angusreidusa.com) United States regulators already care less about whether a chatbot feels futuristic than whether it is transparent, private, and supervised by humans. The Food and Drug Administration updated its guidance on January 6, 2026, and law firms summarizing the change say transparency and human oversight are central to staying outside stricter medical-device scrutiny for some health software. (fda.gov) (natlawreview.com) Privacy is its own minefield. The Federal Trade Commission says companies handling health-related data must honor their privacy promises and maintain security appropriate to the sensitivity of the information, which means a startup cannot casually market intimate health chats as safe without being able to prove it. (ftc.gov) So the real product question is not whether people will pay for an artificial expert. The real question is whether platforms like Onix can label the bot clearly, show what source material it is using, hand off edge cases to a human fast enough, and keep commercial incentives from sneaking into advice that sounds personal. (wired.com)