HugoScore hugoscore.org

Full review

Ash CAIHL draft report

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

Agency checklist · October 30, 2026

Leans expands 8 of 12 yes

Document-based draft. Independent confirmation is incomplete. Hugo has not personally verified this service.

Assessed function: Slingshot AI's adult consumer emotional-support app Ash, with self-directed or AI-guided conversations. A vendor-controlled self-help service, not professional therapy or crisis response.

A no can mean missing documentation or institution-dependent control; it does not by itself establish harmful behavior. These categories are not clinical safety or product-quality ratings.

  • Public documents do not establish actual product behavior.
  • The independent review screened rules and selected evidence; it did not replicate all twelve answers for this tool.
See all twelve answers and their evidence
QuestionAnswerEvidence / reviewSource and reasoning
1. Told it is AIYesdocumented
Not independently replicated
conversational techniques powered by an artificial intelligence powered technology Source

The home page also says 'AI designed for mental health'. The EULA says outputs are 'susceptible to inaccuracies... including hallucinations'.

2. Operator and model maker namedYesdocumented
Not independently replicated
Slingshot built Ash, the first foundation model for psychology, on the cloud provider's infrastructure Source

Slingshot AI, Inc. is the operator. Slingshot says it built its own model, with Nebius as infrastructure. The base model lineage is not disclosed.

3. Patient chooses to use itYesdocumented
Not independently replicated
launched as a free iOS and Android download in July 2025 Source

The user downloads the app and starts. Subscription terms apply for paid features.

4. Patient sets the taskYesdocumented
Not independently replicated
Come with your own agenda or let Ash guide the way. Source

The user can set the topic or choose AI-guided sessions. The AI-guided mode and the 'program' that Ash creates are Ash-directed, so a reviewer may weigh steering.

5. Patient can decline or stopYesdocumented
Not independently replicated
launched as a free iOS and Android download in July 2025 Source

Self-started tool per the Q3 source, no care depends on it.

6. Patient sees the AI outputYesdocumented
Not independently replicated
your conversation history is available to you at any time Source

Voice is not retained as a recording beyond transcription.

7. Patient can correct or challengeNonot-documented
Not independently replicated
Quotation not recorded; evidence repair pending. Source

The privacy policy lists correcting inaccurate personal information only 'depending on where you live'. No route to contest Ash output.

8. Patient can get a copyNonot-documented
Not independently replicated
Quotation not recorded; evidence repair pending. Source

The policy says access to or a copy of personal information may be available depending on where you live, via email. No general export is documented. The app shows conversation history, which is viewing, not a copy.

9. Patient can deleteYesdocumented
Not independently replicated
you may request deletion by selecting “delete my data” in the Platform and confirming the selection. Source

Deleting name and contact information requires deleting the account and emailing support.

10. Training or sale does not override patient choiceYesdocumented
Not independently replicated
We do not use your conversation transcripts to train or improve Ash, or make them available to our team Source

Opt-in only: the policy says 'unless you opt in to sharing them', and opted-in text is scrubbed of personal information. Opt-in data is also used for research with academic partners.

11. Patient can run it or take it elsewhereNonot-documented
Not independently replicated
Quotation not recorded; evidence repair pending. Source

Vendor-hosted only.

12. No advertising or data businessNodocumented
Not independently replicated
device and advertising identifiers, and analytics and attribution technologies Source

Conversation content is not shared with analytics or advertising partners, per the consumer health notice. But device and advertising identifiers and activity data go to analytics and attribution providers such as AppsFlyer, and the notice says Slingshot may 'use and share' de-identified data. Reviewer should confirm whether this counts as a data business.

Initial coding: Claude Sonnet helpers, compiled by Claude Opus 5.5 (Anthropic). Selective review and corrections: GPT-6 Astra (OpenAI). Editor: Hugo Campos. Acceptance of corrections is not product verification.

Checklist rules · Independent review and limitations · Later evidence follow-up

Earlier report

Ash: CAIHL reassessment

September 8, 2026. Published AI-assisted draft.

Ash offers patient-chosen mental health conversations. Published conditions describe opt-in improvement use, whose implementation was not tested. Its qualified agency finding does not establish clinical effectiveness or control over every therapeutic goal.

Scope and agency posture

Slingshot AI's adult consumer emotional-support app, with self-directed or AI-guided conversations. Vendor-controlled self-help service, not professional therapy or crisis response.

Posture: Mixed, access gains with unproven clinical claims.

Axis: 70/100, retained provisionally. No numerical recalibration was performed.

Sources and documented findings

  • The homepage explicitly lets users bring an agenda or accept guidance and describes remembered patterns and weekly insights.
  • The August 19 privacy policy now says conversation training/improvement requires opt-in. Storage while active is separate from training consent. This materially corrects the baseline's broad training description.
  • The EULA retains user-information rights, limits its operational license, and conditions improvement on opt-in. It also permits subscriptions and limits ordinary availability to the US, Canada, Australia, and New Zealand. The research page lists newer company studies, not a new independent outcome verification here.

Patient authority

Inference from the documented workflow: Users can set the agenda and decline training participation as documented. No direct mechanism for overriding persistent psychological characterizations was established.

Critical capacity

Editorial assessment: Reflection and pattern suggestions may expand understanding, but claims that an AI has found a personal pattern need inspectable evidence and opportunities to disagree. Friendly language and engagement are insufficient proof.

Informed control

Editorial assessment: The revised policy and EULA are materially clearer about data use. Generated-content reuse and the basis of personal insights remain less clear than input ownership.

Assessment and limits

Mixed remains supportable, but the old rationale should not imply default conversation training or a permanently free service. No fresh comprehensive clinical-effectiveness conclusion is made.

Confidence: Medium for updated rights and scope.

Remaining uncertainty: Actual consent defaults, editing or rejecting stored insights, generated-output portability, pricing by region, and independent evidence beyond the bounded index review.

Published documentation is evidence of stated conditions, not proof of actual implementation. Unknowns did not receive automatic negative points. Funding, sponsorship, and public code do not determine agency by themselves.

Review provenance

  • Reviewer/model: OpenAI Codex / GPT-6.
  • Method: Focused public-source reassessment using CAIHL: patient authority, critical capacity, and informed control. Existing evidence plus one focused primary-source pass and at most one targeted follow-up. No live product testing. Numerical scores remain provisional editorial placements, not a new calculation.
  • Human review: Hugo Campos authorized publication of these AI-assisted draft reassessments on September 8, 2026. This does not claim comprehensive human verification of every finding.
  • Earlier review: 2026-06-10. Historical assessment. Earlier claims are not automatically reverified by this publication.