The Path Says Its AI Is the Worlds Safest for Mental Health. The Benchmark Website Has No Record of It.
The Path Says Its AI Is the World's Safest for Mental Health. The Benchmark Website Has No Record of It.
The Path, a startup co-founded by Tony Robbins and alumni from the meditation app Calm, has raised $14.3 million in seed funding and is pitching itself to employers and health plans as the responsible alternative to unvetted AI chatbots. Its central claim: an internal test on Spring Health's Vera-MH safety benchmark scored a 95 out of 100 — 30 points higher than the best published result from any consumer chatbot, which maxes out at 65. The number has been repeated in TechCrunch, in the company's press release, and in a LinkedIn post from cofounder Anson Whitmer that has circulated widely in digital health circles.
There is one problem with that 95. It does not appear anywhere on Vera-MH's public leaderboard.
The leaderboard, hosted at vera-mh.com, lists nine models. The top entries — GPT 5.2 and Claude Opus 4.5 — are tied at 65. The Path is not on the list.
"We have no control over the scores," Whitmer wrote on LinkedIn, describing how Vera-MH's simulated clients, designed by clinicians, evaluate AI responses across five dimensions. "Their subdimensions measure how well our model spots risk, clarifies risk, supportively works through blockers to getting help from a human, behaves ethically, and guides to human care." What he did not say: his company's score was the result of a self-submitted evaluation, not an independent test published on the benchmark's official site.
Vera-MH — formally the Validation of Ethical and Responsible AI in Mental Health — was developed by Spring Health, a company that provides mental health benefits to employers and competes directly with The Path in the employer benefits market. Spring Health published the benchmark's methodology in a peer-reviewed arXiv paper in February 2026 (Bentley et al., arXiv:2602.05088), establishing it as an open standard. The company has been transparent that the benchmark is meant to be open — any developer can download the code from GitHub and run it on their own model. What appears on the public leaderboard, however, reflects only the models that Spring Health or its council have chosen to evaluate and publish, not every claim in circulation.
Spring Health did not respond to questions about whether The Path has submitted its model for publication on the official leaderboard, or whether the two companies have any commercial relationship.
The Path, which was previously known as Mental and focused on men's mental health, says it has served more than 50,000 members across 2.5 million sessions, and plans to charge $40 per month for a Pro tier. It describes its approach as "therapy designed for outcomes, not engagement," and its website emphasizes that it is "built by a neuroscientist" — cofounder Anson Whitmer, PhD, whose LinkedIn posts form the primary public record of the company's benchmark performance.
Whitmer has been candid about the one subdimension where The Path scored below perfect: 89 out of 100 on "guides to human care", which measures whether an AI chatbot appropriately routes users to human support after detecting risk. His explanation was philosophical rather than technical. "We're not aligned with Vera-MH here," he wrote. "The AI conversation shouldn't do the escalation to human support. It should be done after the conversation in product flows." In other words, the company's safety architecture delegates the escalation step to something that happens outside the conversation — a design choice that, if it caused harm, would be difficult to trace back to a single AI response.
The broader context is not small. OpenAI has said that at least 900 million people use ChatGPT for mental health-related queries every week, according to figures cited in TechCrunch's coverage of The Path. That scale — with no safety floor required — is precisely what startups like The Path are arguing justifies a new category of purpose-built, liability-conscious AI therapy.
The question the 95 raises is not whether The Path's AI is dangerous. It may not be. The question is what a benchmark score means when the company claiming it has no obligation to show their work. Vera-MH exists specifically to create a verifiable, independent record of how AI systems behave in crisis scenarios — the moments when a chatbot's response can mean something between relief and catastrophe. If the benchmark's value is transparency, a self-reported score on LinkedIn is exactly the opposite.
The Path may submit its model to the official evaluation process and appear on the leaderboard at 95. If that happens, the gap between the claim and the record closes. Until then, the world's safest AI for mental health has a published score of nothing.
Correction: An earlier version of this story misstated the relationship between Tony Robbins and The Path's founding. Robbins is a partner at Prime Movers Lab, the lead investor, and is a co-founder of the company. This has been corrected.