Mental health AI
Ash
Ash is a free consumer mental health chatbot from Slingshot AI, a startup with $93 million raised, launched in July 2025 as 'the first AI designed for therapy' and now marketed as 'AI designed for mental health' with explicit disclaimers that it is not for crisis use and not a replacement for professional care. The company reports over 150,000 users and publishes its own real-world outcome and safety studies, which independent experts quoted by STAT found insufficient for efficacy claims, and it pulled the app from the UK in January 2026 after concluding there was no clear regulatory pathway under medical device rules. The CAIHL read: strongly patient-chosen in use, but the boundary between therapy and wellness positioning, the company-controlled evidence base, and conversation-data training make alignment a live question rather than a settled one.
Public-source research has been drafted; final human publication review and change-log detail are still required.
Summary judgment · 70% toward patient-directed
Mixed, access gains with unproven clinical claims
Ash is freely chosen and controlled by its adult users, which is genuinely patient-directed, but the vendor controls the model and the evidence narrative, conversations feed model improvement, efficacy claims rest on company-run uncontrolled studies, and the product was withdrawn from the UK in January 2026 over medical device regulation concerns.
Patient agency
How this tool changes agency
Ash is built for reflection, pattern insight, and everyday emotional processing, and company studies report depression and anxiety improvements, but it is explicitly not designed for crisis, does not navigate the care system, and its outcome claims lack controlled or independent evidence.
The privacy policy grants access, correction, deletion, and objection rights over personal data, and a delete-my-data button exists, but no mechanism is disclosed for challenging the AI's psychological insights or pattern claims about a user beyond emailing support.
Patient-facing signals
Who does this AI serve?
Users choose Ash freely and pay nothing, but the venture-funded vendor controls the model, uses scrubbed conversation text to improve it, and has repositioned its clinical language between therapy and wellness framings as regulatory pressure shifted.
Can patients tell AI is involved?
Ash is openly marketed as an AI, the product is the AI conversation itself, and the site discloses that it is not a human therapist and not for crisis use.
Can patients meaningfully choose?
Use is voluntary, free, and consumer-initiated with an in-app delete-my-data option, though the app is restricted to adults 18 and older and the privacy page's claim that conversations are not stored unless users opt in sits uneasily with policy text saying conversation data is retained while the account is active.
Can patients correct or challenge what the AI produces?
The privacy policy grants access, correction, deletion, and objection rights over personal data, and a delete-my-data button exists, but no mechanism is disclosed for challenging the AI's psychological insights or pattern claims about a user beyond emailing support.
Does it help patients understand or act?
Ash is built for reflection, pattern insight, and everyday emotional processing, and company studies report depression and anxiety improvements, but it is explicitly not designed for crisis, does not navigate the care system, and its outcome claims lack controlled or independent evidence.
Text findings
Who is left out or burdened?
Adults-only, crisis-excluded, and geography-dependent
Under-18 users are excluded, people in crisis are redirected to outside resources, language and accessibility support are not disclosed, and UK users lost access entirely in January 2026, after which the company's own analysis said most turned to general-purpose AI or to no support at all.
What happens to patient data?
Meaningful disclosure with a training-use core
Conversation text is collected, scrubbed of personal identifiers, and used to train and improve the model, voice audio is not stored, data is encrypted with coded identifiers, deletion is available in-app, and data may go to service providers, affiliates, and law enforcement; retention lasts while the account is active, marketing copy and policy text differ on opt-in storage, and no HIPAA status is disclosed.
Are the clinical boundaries clear?
Partial and shifting
The site now says Ash is not for crisis and points to findahelpline.com, and FAQs address professional-help limits, but the product launched as 'the first AI for therapy,' later softened to mental health wording, holds no FDA authorization, and withdrew from the UK because its category under medical device rules was unresolved.
Who defined what good looks like?
Vendor-defined, with named advisors
Outcome and safety studies are run and published by Slingshot itself, including a 300-person single-arm study and internal safety audits, with a prominent advisory board, while STAT reported independent experts found the evidence could not support efficacy claims; no independent controlled trial was found.
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.
Draft profile · Medium draft, official sources and credible press, clinical and safety claims rest mainly on company-run studies