HugoScore hugoscore.org

Full review

Ash CAIHL draft report

Evidence-linked HugoScore draft report for a health AI tool that affects patients.

Back to profile Markdown source

HugoScore CAIHL Draft Report: Ash

Status: Draft for human review Last reviewed: 2026-06-10 Review method: Public-source review of the official website, privacy policy, clinical-research page, launch press release, and STAT and Behavioral Health Business reporting; no app walkthrough, full EULA review, vendor interview, or independent model evaluation. Service: Ash Vendor: Slingshot AI, Inc. Category: Mental health AI

Summary

Ash is a free consumer mental health chatbot from Slingshot AI, a startup that has raised $93 million from investors including Andreessen Horowitz, Radical Ventures, and Forerunner. It launched in July 2025 billed as "the first AI designed for therapy," trained on what the company describes as a proprietary dataset of clinically relevant behavioral health data, with a named clinical team and advisory board. The current site markets it as "AI designed for mental health" for adults 18 and older, reports more than 150,000 users, and carries a standing disclaimer that Ash is not designed for crisis use, pointing users to findahelpline.com. In January 2026 Slingshot withdrew Ash from the UK, saying there was no clear regulatory pathway for wellbeing products under medical device rules.

From a CAIHL perspective, Ash is genuinely patient-chosen: free, voluntary, consumer-controlled, and transparent about being an AI. The alignment question is harder. The vendor controls the model and publishes its own evidence, including a 300-person single-arm study that independent experts quoted by STAT said cannot support efficacy claims. Conversation text, scrubbed of identifiers, is used to improve the model. And the product's positioning has moved between therapy and wellness language as regulatory pressure shifted, which matters because the wording determines which safety rules apply to a tool that vulnerable people use for mental health.

Evidence Reviewed

CAIHL Profile

  • Who does this AI serve? Patient-facing and patient-chosen, vendor-aligned commercially. Users pay nothing and choose freely, while the venture-funded vendor controls the model, the data, and the evidence narrative.
  • Can patients tell AI is involved? Yes. The AI is the product, and the site says plainly it is not a human and not for crisis.
  • Can patients meaningfully choose? Yes, for adults. Use is voluntary and free with in-app data deletion, though under-18s are excluded and the data-storage opt-in language is inconsistent between the privacy summary and the policy text.
  • Can patients correct or challenge what the AI produces? Partial. Data rights cover access, correction, and deletion, but there is no disclosed way to contest the AI's psychological insights or pattern claims beyond emailing support.
  • Does it help patients understand or act? Partial. It supports reflection, insight, and everyday coping, with company-reported symptom improvements, but it does not handle crises, does not navigate care, and its outcomes lack independent or controlled evidence.

Agency Interpretation

Ash's clearest agency value is a free, always-available, judgment-free space that adults choose for themselves. For people who cannot afford or reach therapy, or who are deterred by stigma, that is a real expansion of options, and the company has invested more visibly in clinical staffing, safety auditing, and research publication than most consumer chatbot vendors.

The unresolved CAIHL tension is who gets to define what this product is and whether it works. Slingshot launched with therapy language, softened to wellness language, and exited the UK rather than be classified as a medical device. Its efficacy and safety evidence is self-run and self-published, and STAT reported that outside researchers found the flagship study could not support its claims. Meanwhile user conversations train the model. Patients get the benefits of a purpose-built mental health AI, but the category it occupies, the rules that bind it, and the proof that it helps all remain under vendor control. For a product used by people in distress, that asymmetry is the central concern.

Key Unknowns

  • Whether any independent randomized or controlled evaluation of Ash exists or is planned. None was found.
  • How in-app crisis detection actually behaves, including what triggers escalation language and how reliably it appears. Company audits describe this, but no independent test was found.
  • How the opt-in conversation storage setting works in practice, and what the default is, given the conflict between the privacy summary and the policy text.
  • Whether Ash has been named in any FTC inquiry, state enforcement action, or attorney general complaint. None naming Ash specifically was found in this review, though the broader AI therapy category is under active FTC and state scrutiny.
  • What the proprietary training dataset of behavioral health data consists of and whose consent covered it.
  • Monetization plans, since the app is currently free and venture-funded with no disclosed revenue model.

Publication Recommendation

Ready for human review as a draft profile. Keep confidence at medium and revisit after any independent evaluation, FTC or state action, UK re-entry, or monetization change. An app walkthrough to verify crisis handling and the data opt-in default would raise confidence most.