Anistratenco Health Assistant CAIHL draft report

Anistratenco Health Assistant: CAIHL review

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.

• S1: Health Assistant homepage — Anistratenco. Current assistant and future connection roadmap.

• 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.