HugoScore hugoscore.org

Full review

Anistratenco Health Assistant CAIHL draft report

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

Agency checklist · October 30, 2026

Leans expands 9 of 12 yes

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

Assessed function: Current consumer Health Assistant / HealthOS web offering. Future direct health connections, diagnosis and prescribing are excluded.

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
Anistratenco Health Assistant is an AI health assistant that helps you explore health questions Source

Home page FAQ. Terms section 'AI outputs and research' also warns outputs can be wrong.

2. Operator and model maker namedNonot-documented
Not independently replicated
Quotation not recorded; evidence repair pending.

Operator is named as Anistratenco Platform, LLC, a Delaware LLC (Privacy and Terms, effective 7 Sept 2026), which supplies the legal operator the earlier review could not find. No AI model or model maker is named. Providers are described only by category.

3. Patient chooses to use itYesdocumented
Not independently replicated
Open Health Assistant and share what is on your mind. Source

Direct-to-consumer web app with a paid Plus or Pro plan. The user signs up and starts.

4. Patient sets the taskYesdocumented
Not independently replicated
Ask away about symptoms, medications, supplements, sleep, nutrition, training, recovery, or a result you want to understand. Source

The user sets the question. Answer style settings and instructions are also user-set.

5. Patient can decline or stopYesdocumented
Not independently replicated
Open Health Assistant and share what is on your mind. Source

Self-started tool rule: Q3 is yes and no care depends on it. Source is the Q3 source.

6. Patient sees the AI outputYesdocumented
Not independently replicated
Ask a follow-up, open the cited sources, and correct anything that is wrong. Source

The chat answer, sources and saved memory are shown in the app.

7. Patient can correct or challengeYesdocumented
Not independently replicated
See what the Assistant remembers. Correct it or remove what no longer belongs. Source

Memory history and the health avatar are editable ('Review it, correct it'). Chat answers can be challenged in a follow-up. The privacy policy also offers correction requests by email.

8. Patient can get a copyYesdocumented
Not independently replicated
export an encrypted local backup, download supported account records and cloud-encrypted files on a trusted device Source

Privacy policy consumer-health section. Account export is also described under 'Your controls and privacy rights'.

9. Patient can deleteYesdocumented
Not independently replicated
use account privacy controls or support to request deletion of eligible managed account information Source

Home page FAQ. Browser-local data can be cleared by the user. Billing, security and backup records are kept for set periods.

10. Training or sale does not override patient choiceNodocumented
Not independently replicated
Other providers process requests under their published retention and training terms; we do not describe those routes as zero-data-retention or contractually no-training. Source

The operator says it does 'not currently use raw health conversations or files to train its own models', but model providers' training terms are not controlled or documented, and no opt-out is given.

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

Vendor-hosted web service. Export is documented but no way to run it yourself or move the function with your data to another provider without the operator.

12. No advertising or data businessYesdocumented
Not independently replicated
We do not share it with advertisers, data brokers, employers, insurers, affiliates, or another third party for that party's own purpose. Source

Privacy policy also says it does not sell consumer health data or share raw health content for targeted advertising. Model providers are processors for the feature. De-identified information may be kept for statistics and security.

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

Anistratenco Health Assistant: CAIHL review

Method v1.3 review, October 30, 2026

Agency finding: Unclear. Reason: benefit documented, constraint search incomplete.

Control: No position: model provider not disclosed. Toward the patient-directed end on documents, short of it. The patient sets the question, the form of help and standing instructions, can change direction, can ask for evidence, assumptions and competing explanations, and owns the output. No institutional rail on the answer content is documented: no sponsor, no clinical service and no product sales. The operator keeps a necessary power: it owns provider routing and can change the model at will, so full workflow authority stays with Anistratenco. Every condition rests on operator documentation and untested behavior. Documentary evidence alone points to one position, which is withheld pending protocol sessions under rubric section 8.

Disclosure (transparency gate): Service operator named: yes, in public documents. Model developer named: not found in public documents. AI names its operator when asked: not tested (documentary review).

Evidence: Documentary review. Public sources only, no hands-on testing by HugoScore.

Review status: Human adjudication: Hugo Campos settled named questions (what was reviewed, the disclosure results, any coder disagreement). Not verification of the service's claims.

Coded by: Claude Opus 5.5 (Anthropic), coder 1, and GPT-6 Astra (OpenAI), coder 2, coded independently. Conflict check open: the model developer is not known.

What it helps a patient do

  • Interpret one's own records with sources (Understand, documented, not tested). The patient can upload labs, letters and files and ask for explanations with cited sources and visible uncertainty, and can ask which study supports a conclusion (E10, E12, E24). Comparator: the patient's realistic alternative is searching the web or waiting to ask a clinician; the operator's documents describe the function, not a measured advantage. Operation untested.
  • Competing explanations on request (Ask better questions, documented, not tested). The 'Epistemic' and 'Curious' answer forms are documented to explain competing explanations and explore alternatives (E11). Operation untested.
  • Prepare questions and next steps (Act, documented, not tested). The docs describe asking the assistant to organize questions for an appointment and to list 'which questions would be useful to ask next', and the homepage offers 'Summarize this into a next-step plan' (E10, E12). Operation untested.

What limits the patient

  • None supported in this review.

Strongest evidence against these findings

Control: every C condition rests on Anistratenco's own help pages and settings descriptions (E10, E11), which are claims, not observed behavior; a session could show the assistant steering, refusing alternatives or ignoring instructions, which would move C2 or C3 to N. The T2 code rests on a literal reading of the routing clause; read strictly, no model developer is disclosed, the gate fails with 'model provider not disclosed', and no position or withheld position is shown. Posture: the unnamed providers 'under their published retention and training terms' (E4) may mean that health content can be used to train third-party models, which would make the notice contradiction with E13 a supported constraint and the posture Mixed. In the other direction, if the in-product disclosure names the providers and their no-training terms, the candidate dissolves and the posture becomes Potentially agency-expanding.

Open questions

  • Names of model providers and inference hosts, per route and plan (T2 under the strict reading, conflict check)
  • Which routes, if any, permit provider training on health content
  • In-product feature-specific disclosures and consent screens
  • Whether behavior matches the documented answer forms and redirection
  • Operator registration outside the product
  • Governing law and forum shown at checkout

What changed from the earlier review

Posture moves to Unclear because v1.3 requires a completed constraint search for Potentially agency-expanding, and the unnamed providers' training terms cannot be read while a notice contradiction of unresolved scope remains. The control conditions now resolve on documents to a withheld position, subject to the T2 reading. Coordinator correction under F28: T2 is not found, so the transparency gate withholds any position (model provider not disclosed).

Sources

Earlier review, September 28, 2026

The review below was prepared on September 28, 2026 with OpenAI Codex / GPT-6 (exact variant not recorded), before method v1.3. It is kept unchanged as history. Where it differs from the v1.3 review above, the v1.3 review is current.

Reviewed September 28, 2026. AI-assisted draft. Potentially agency-expanding; unscored.

Executive summary

A patient-initiated health assistant with cited research and local-context controls described by the vendor. Potentially agency-expanding, but provider-specific processing terms and untested correction behavior limit confidence. Unscored.

CAIHL question: who does this AI serve?

Patient-initiated consumer health AI; vendor-operated service. Primary user: Individuals asking health questions using their chosen context. Browser-local context and encrypted synchronized files are claimed; AI processing can involve external providers. Consumer choice supports a patient-directed pathway, but does not by itself establish patient alignment. Vendor interests in subscriptions and continued use coexist with the patient's purposes.

What the service does and the scope of this review

Current consumer Health Assistant / HealthOS web offering. Future direct health connections, diagnosis and prescribing are excluded.

Patient-impact pathway

AI interpretation of health information can shape questions, understanding and self-management decisions.

Mixed HugoScore profile

Who sets and changes the goal?

Partial. Consumer users choose questions and context; persistence of corrections or refusal to follow vendor framing was not tested. S1 S5

Can patients tell AI is involved?

Yes. The public offering explicitly describes health AI reasoning. S1 S5

Can patients meaningfully choose?

Partial. Voluntary consumer access and deletion/export routes are described; complete exit and provider-copy deletion were not verified. S2 S3

Can patients correct or challenge the output?

Not tested. Data rights and user input ownership do not establish reliable correction of generated interpretation or memory. S2 S3

Does it help patients understand or act?

Potentially. Cited research and chosen health questions may support reflection and visit preparation; advocacy and citation fidelity were not tested. S1 S5

Can patients evaluate the conditions of use?

Partial. Storage and processing are distinguished, but active provider/model identities and their effective terms were not established. S2 S4

What happens to patient data?

Conditional provider protections. The policy describes encrypted local context and cloud ciphertext. Chosen analysis can require plaintext processing. No-storage requests depend on provider support; provider retention/training terms can apply. No own-model training of raw health conversations/files is stated. [S2, S4]

Who is left out or burdened?

Paid and metered. Monthly plans show Plus $19 and Pro $49 with compute allowances. Digital access, affordability, accessibility and subgroup performance remain unevaluated. [S3, S5]

Are the clinical boundaries clear?

Documented; behavior untested. Terms limit the current service to consumer informational assistance, excluding diagnosis, prescribing, emergency response and continuous monitoring. Citations can be wrong. [S3]

Who defined what good looks like?

Not established. The inspected sources do not establish patient partnership in defining evaluation outcomes or independent validation of the current assistant.

Current product versus roadmap

The homepage describes direct health connections as planned, with a Health space target of October 7, 2026. They receive no current capability credit. The chat entry was observed behind sign-in; no response or correction behavior was tested. Local storage, synchronized storage and external processing are separate claims. The security page explicitly disclaims an audit, certification or BAA. Export and deletion are described, not independently verified. Homepage · Security · Chat

Key unknowns

  • Legal operator identity in the inspected public policies
  • Active models, providers and applicable retention/training settings
  • Ability to correct persistent inferred context and challenge generated advice
  • Citation accuracy, escalation behavior and independent product evaluation
  • Accessible use and subgroup performance

Risks to investigate

These are plausible failure modes, not observed harms.

  • Mistaking encrypted storage for absence of external plaintext processing
  • Overreliance on cited but unverified interpretations
  • Loss of local information without recoverable keys/backups; vendor describes this limitation
  • Cost or usage limits interrupting a chosen workflow

Patient agency interpretation

May help people interpret their own information, inspect sources and prepare questions they choose. External processing, metered access and unverified correction of inferred context can limit informed control. Critical reflection requires inspectable evidence; strategic action requires the ability to use the result for the patient's own purposes; neither is proved by engagement or vendor assurances alone. Chosen questions, inspectable sources and record control could support patient authorship. This is a conditional qualitative judgment, not demonstrated alignment or a numerical ranking.

Publication recommendation

Publish as a source-backed AI-assisted draft. Retain a null numerical axis position: documented architecture is insufficient for a calibrated comparative number. Confidence: Medium for published terms, low for practical agency and behavior (AI-assisted draft).

Evidence ledger

All sources accessed September 28, 2026. Vendor disclosures establish what is promised, not that controls work.

  • S2: Privacy — Anistratenco. Effective September 7, 2026; local storage, processing, provider terms and rights.
  • S3: Terms — Anistratenco. Effective September 7, 2026; consumer limits, ownership and subscription usage.
  • S4: Security — Anistratenco. Reviewed September 7, 2026; architectural claims and assurance limits.
  • S5: Plans — Anistratenco. Monthly Plus and Pro; citations and bounded compute.
  • S6: Chat entry — Anistratenco. Observed sign-in gate; no conversation tested.

Review provenance

criteriaSource: CAIHL-derived HugoScore framework and current qualitative review method

reviewer: AI-assisted public-source review prepared in OpenAI Codex

model: OpenAI Codex / GPT-6 (exact variant not recorded)

humanReview: Hugo Campos authorized review and publication. No comprehensive human verification is claimed.

reviewDate: 2026-09-28

limitations: Public-source documentary review only. No account creation, health-data upload, interaction tests, security audit, clinical evaluation or vendor interview. Host-specific and provider-specific behavior is unverified.