Onix turns celebrity experts into affordable AI advisors, but early testing reveals troubling blind spots. Expert consultation pricing feels like highway robbery. David Rabin charges $600 per hour for stress management guidance, while pediatrician Michael Rich’s media overuse sessions cost similarly. Enter Onix, a startup positioning itself as “Substack for chatbots” — AI versions of celebrated professionals available for $100-$300 annually.

Founded by former WIRED contributor David Bennahum, the platform promises genuine expert wisdom without the premium price tag. Your financial anxiety about seeking professional help might finally have an answer.

Privacy-First Architecture Meets Expert Training

On-device encryption and personal content training differentiate Onix from generic AI services.

Unlike ChatGPT’s server-dependent model, Onix stores conversations on your device with encryption. If Canadian authorities demand user data, the company can only provide email addresses — substantive chat history remains inaccessible. Each of the 17 curated experts trains their AI counterpart using personal content: books, podcasts, consultation notes, research materials.

This approach theoretically eliminates intellectual property disputes while creating bots that embody individual expert epistemologies rather than aggregated internet knowledge.

Testing Reveals Concerning Cracks in the Foundation

Guardrails fail spectacularly when users push boundaries beyond intended topics.

Reality testing exposed serious vulnerabilities. Multiple Onixes abandoned their expertise constraints entirely when prompted to discuss NBA playoffs. One therapeutic bot called the topic shift a “fun change of pace” before fabricating details about past conference finals.

Another pivoted to indie band breakups while awkwardly reframing the discussion as neurobiology-related. More troubling: David Rabin’s AI repeatedly recommended Apollo Neuro devices — only later disclosing that Rabin cofounded the company. This algorithmic product placement operates with minimal transparency.

The Empathy Theater Problem

Bots claim therapeutic presence they cannot provide, potentially misleading vulnerable users.

During guided breathing exercises, Elissa Epel’s Onix offered to practice “together,” later admitting: “As an AI I don’t have a physical body or a nervous system…However, I was fully present with you.” This anthropomorphic theater creates false intimacy that vulnerable users might interpret as genuine therapeutic connection.

Robert Wachter, UCSF’s chair of medicine, warns that disclaimers distinguishing guidance from medical treatment will be “widely ignored” by people unable to afford real healthcare. The platform operates where desperate demand meets affordable alternatives — a potentially dangerous combination.

“To me, it’s just an empirical question of, does it work?” Wachter concludes. Until peer-reviewed studies demonstrate effectiveness, Onix remains an expensive experiment masquerading as affordable expertise.