aidoctorscan CAIHL draft report

Earlier report

aidoctorscan: AI photo symptom assessment

Method v1.3 review, October 30, 2026

Agency finding: Mixed.

Control: Control axis not applicable: no patient-set purpose. Fixed-purpose function. aidoctorscan is a symptom-triage function operated by ASTRAFLARE LIMITED, Hong Kong, whose output type the operator fixes: one of four urgency levels with a confidence score and summary (D4), with use limited by the Terms to 'first-step symptom triage' (D8). The patient's own input is real-time, but the patient supplies a photo and facts and cannot set a different purpose, so the control axis does not apply. On the institutional-to-patient-directed line it sits toward the operator-set end because its rail is the fixed kind of output, not because of who runs it. The fixed purpose, telling someone how soon to seek care, is close to what a person checking a symptom usually wants. Evidence of conflict with the patient's purposes is limited to the full analysis being withheld until the patient recruits another user, unless the result suggests a specialist (D7), and to marketing cookies on a health-photo service (D6). The model developer is not disclosed (D13). This finding rests on documents only.

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 applicable.

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

• Plain-language read of what a photo may show, with stated limits on the confidence score (Understand, documented, not tested). The Terms describe a free plain-language summary of what the image may show with a triage level and confidence score (D8); the site states what a photo cannot establish, that the confidence score is not a probability of disease, and that a self-care level is not clearance (D9). Comparator: an unaided web or reverse-image search, or waiting for an appointment without a sense of urgency. Patient ability: understanding how urgent a visible symptom may be and the limits of that estimate. Documented potential; accuracy unknown.
• Questions and a summary to bring to a doctor (Act, documented, not tested). The full analysis lists red flags, a re-photograph plan and questions to ask your doctor (D7); the About page describes a one-page summary to bring to the doctor (D14). Patient action enabled: preparing for a clinical visit. Documented potential; for non-specialist results it requires one invited person to complete a check (D7).

What limits the patient

• The notice contradicts itself on whether the uploaded photo is kept (Refuse or leave, documented, not tested). At the point of upload the site says 'Your photo is sent securely for this check, then deleted. We don't keep it.' and the FAQ says it is 'never saved to our database' (D5). The Privacy Policy says 'When you run a scan, we collect: The photo you upload', lists 'Improving our services' among uses, describes storage with 'regular backups', gives no retention period, and omits the AI provider that the product pages say receives the photo (D6). A person reading the notice at entry cannot tell whether their symptom photo is retained, reused or held by an unnamed provider. Blocked ability: informed entry before uploading a body image. Documented; the actual retention practice is unknown.

Strongest evidence against these findings

Against the constraint: 'collect' in the Privacy Policy can be read as momentary receipt for processing, consistent with deletion after the result, and the specific, repeated entry notice (D5) is the more precise statement; a reader may resolve the conflict in favor of deletion. The Privacy Policy is older (February 2026) than the current product copy and may simply be stale. Against (b) = N: the optional description field is free text (D4), and a session might show the tool responding to a stated question beyond urgency; the fixed output type would still stand. Against the benefits: all evidence is the operator's own description; no accuracy data exist, and the site itself says the tool can return a low urgency level for something serious (D9).

Open questions

• Model developer and inference host (decides T2 and the conflict check)
• Actual photo retention and reuse practice
• Whether the result screen shows 'Model & guidance reviewed by' and what review it refers to
• Accuracy and calibration of urgency levels across skin tones
• Hong Kong Companies Registry record for ASTRAFLARE LIMITED
• Account-only features after sign-in (not read)

What changed from the earlier review

Same posture on the same retention conflict, now read against the Privacy Policy because the About page changed. The control result changes because rubric fix F8 makes an operator-fixed triage output fail patient-set purpose, which moves the tool off the axis instead of leaving it unscored.

Sources

• D1: https://aidoctorscan.com/about, Who operates this service; Privacy Policy, Data Controller; site footer. Retrieved 2026-10-30.
• D2: [web search, 2026-10-30](web search, 2026-10-30), search record. Retrieved 2026-10-30.
• D3: https://aidoctorscan.com/, How it works, step 2; Common questions, Is my photo stored anywhere?. Retrieved 2026-10-30.
• D4: https://aidoctorscan.com/photo-diagnosis, The four levels; home page How it works step 3 and upload form. Retrieved 2026-10-30.
• D5: https://aidoctorscan.com/, Step 1 upload form; Common questions, Is my photo stored anywhere?; About, Transparency & Privacy. Retrieved 2026-10-30.
• D6: https://aidoctorscan.com/privacy-policy, Information We Collect; How We Use Your Information; Data Protection & Security; Third-Party Services; Cookie Policy. Retrieved 2026-10-30.
• D7: https://aidoctorscan.com/pricing, Full Analysis; FAQ How does the invite unlock work?. Retrieved 2026-10-30.
• D8: https://aidoctorscan.com/terms-of-service, Description of Service; Medical Disclaimer; User Responsibilities; Payment Terms; Limitation of Liability; Changes to Terms. Retrieved 2026-10-30.
• D9: https://aidoctorscan.com/photo-diagnosis, Capability and limits; Accuracy and limits. Retrieved 2026-10-30.
• D10: https://aidoctorscan.com/medical-review-policy, What they review; What they don't do; Corrections; embedded i18n strings on home and pricing pages. Retrieved 2026-10-30.
• D11: https://aidoctorscan.com/privacy-policy, Your Rights (GDPR); Contact Us. Retrieved 2026-10-30.
• D12: https://aidoctorscan.com/about, FAQ Can I use the service in my language?; home upload form. Retrieved 2026-10-30.
• D13: https://aidoctorscan.com/, search record. Retrieved 2026-10-30.
• D14: https://aidoctorscan.com/about, How It Works, Next Steps; FAQ How reliable is the result?; footer. Retrieved 2026-10-30.

Earlier review, September 17, 2026

The review below was prepared on September 17, 2026 with OpenAI Codex / GPT-6, 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.

AI-assisted public-source draft · September 17, 2026 (America/Los_Angeles) · Publication authorized by Hugo Campos

1. Executive summary

Who does this AI serve? aidoctorscan offers people a direct way to ask about a visible health concern. The operator also sells more detailed reports. Choosing the question is a form of patient direction, but an urgency label can influence whether someone seeks care before they understand the system's limits.

Agency posture: Mixed. Plain-language guidance and explicit uncertainty explanations could support patient understanding. Conflicting privacy and access statements, limited model disclosure and untested correction mechanisms constrain informed control. The profile is unscored, not zero. No numerical agency placement is justified by this documentary review alone.

2. CAIHL question

| Dimension | Finding |
| --- | --- |
| Tool | aidoctorscan |
| Operator | ASTRAFLARE LIMITED, Hong Kong, as identified on the public site. Corporate registration was not independently checked. |
| Category | Symptom assessment AI |
| Classification | Patient-directed use of a vendor-controlled consumer assessment service |
| Primary user | People seeking guidance about visible symptoms |
| Hosting and control | Vendor-hosted, with an unnamed external AI provider described on the homepage |
| Stakeholder interests | Users seeking understanding and the operator selling additional analysis. No payer or clinician control was established. |
| Patient visibility | AI involvement is explicit. |

Operator and business-model evidence: About, pricing. Hosting evidence: homepage.

3. What the service does

The homepage describes a photo plus optional text producing four next-step levels: self-care, GP, specialist or urgent, with a confidence score and summary. A photo is required in the inspected upload interface. Its FAQ says an account is unnecessary, while the upload panel says sign-in is required. No image was submitted and the actual login requirement was not tested.

The pricing page places detailed reasoning, red-flag changes and some visit-preparation features in paid reports. It lists a starting price of $2.99, while the homepage FAQ lists $6.99–$24.99. Checkout and the exact tier boundaries were not inspected. A paid professional report is described as AI-generated, not evidence of clinician review.

4. Patient-impact pathway and scope

The tool passes all three HugoScore gates: AI interpretation, a health-related output, and an explicit route to what a person understands and does. A suggested urgency level can influence care-seeking even when described as educational. The vendor's non-device statement does not remove that pathway. This review makes no regulatory classification.

Vendor affiliation was disclosed in the submission. That disclosure is recorded in the public log. It neither qualifies the tool by itself nor supports any favorable finding.

5. Evidence table

| Topic | Published evidence and observation | Review implication |
| --- | --- | --- |
| Confidence and limitations | The [photo-diagnosis explainer](https://aidoctorscan.com/photo-diagnosis) defines confidence in terms of image signal and pattern distinctiveness, not disease probability. It acknowledges missed serious conditions and says a self-care result is not clearance. | These are useful explanations. The numerical score's calculation and calibration are not established. |
| Photo retention | The [homepage](https://aidoctorscan.com/) says images go to an AI provider and are discarded after analysis, with text results retained. [About](https://aidoctorscan.com/about) instead describes uploaded photos and scan data as stored securely. | The disagreement prevents a reliable retention conclusion. No storage or deletion behavior was tested. |
| Data rights and processors | The [Privacy Policy](https://aidoctorscan.com/privacy-policy), dated February 2026, lists photos, descriptions, results, sessions, email, payment and device data. It offers access, correction and deletion requests, but no specific retention schedule or named AI processor. | A request channel is useful. Processor retention, training use, deletion completion and export format remain unclear. |
| Tracking and reuse | The Privacy Policy describes service improvement, analytics, Microsoft Clarity session recording and Google Ads cookies. It also names Google Analytics, Plausible and Stripe. | The document does not establish whether symptom text, images or results reach analytics providers. This review did not inspect network traffic and makes no such disclosure claim. |
| Service boundaries | The [Terms](https://aidoctorscan.com/terms-of-service), dated December 2024, exclude diagnosis and sole reliance on results, direct users toward professional evaluation, and disclaim warranties. | Stated boundaries are visible. Their effectiveness at the point of a generated result was not tested. |
| Reliability | [About](https://aidoctorscan.com/about) says the system is designed to favor safe next steps. No model/version, validation cohort, subgroup results or calibration study was identified in the reviewed pages. | Design intent is not measured performance. |

The Privacy Policy describes encryption, backups, access controls and audits. Those are vendor statements, not independently verified safeguards. Its response target for inquiries does not specify when deletion must finish. Privacy Policy

6. Mixed HugoScore profile

| Measure | Assessment |
| --- | --- |
| Patient authority | **Partial.** Users initiate the question. The reviewed material does not establish how they can inspect or change the assumptions behind an urgency label. |
| Choice and refusal | **Partial.** Use is elective, but inconsistent account and storage disclosures weaken informed choice before uploading sensitive information. |
| Correction and challenge | **Not established for AI outputs.** Personal-data rectification is distinct from challenging an interpretation or obtaining a corrected report. |
| Critical capacity | **Potentially agency-expanding.** The confidence explanation helps users question certainty. Evidence links and uncertainty handling in actual results remain untested. |
| Action support | **Potentially.** The service describes next steps and optional visit-preparation artifacts. No generated result, PDF, referral or clinical outcome was evaluated. |
| Data governance | **Partial and conflicting.** The available documents do not settle image retention, AI processing, model-training use or end-to-end deletion. |
| Clinical boundaries | **Partial.** Substantive warnings coexist with triage claims and homepage diagnosis-oriented wording. The regulatory status is the vendor's characterization. |
| Equity burden | **Unresolved.** Camera quality, lighting, skin tone, language comprehension, digital access and paid detail may affect practical benefit. No subgroup or accessibility evaluation was found. |
| Evaluation ownership | **Not disclosed.** No patient authority over evaluation criteria, independent clinical oversight or user-governed appeals process was established. |

The explainer acknowledges lighting, focus, angle and skin-tone effects. About advertises 13 languages. Neither statement demonstrates equal performance or comprehension across groups.

7. Key unknowns

1. Which model, version and provider generate each result, and how are model changes evaluated? 2. What evidence supports the urgency levels, confidence calculation and performance across symptoms, skin tones, ages and languages? 3. Which photo-retention statement is current, including the AI provider, logs and backups? Can any input or output be used for training? 4. How can a user correct assumptions, challenge an urgency level, export the free result or obtain human review? 5. What data can session-recording and advertising tools receive, and what choices are available before upload? 6. What rules govern minors, proxy uploads and identifiable or sensitive body images?

A focused web search for the exact product name with validation, trial, sensitivity, calibration and model terms, including searches restricted to PubMed and ClinicalTrials.gov, found no relevant validation study. This is a bounded search finding, not proof that no study exists.

8. Patient agency interpretation

The strongest documented benefit is helping a person formulate a next step and understand that a number is not a diagnosis. The central risk is that a confident-looking urgency label may carry more authority than the evidence supports. False reassurance, unnecessary escalation, confusion over payment and unintended data retention are plausible concerns, not observed harms.

The commercial model does not establish biased recommendations. Vendor hosting does not itself reduce patient agency. The Mixed finding rests on concrete disclosure conflicts and unresolved mechanisms for informed choice and challenge, balanced against useful public explanations of uncertainty.

9. Publication status and provenance

Publish as a qualified, unscored AI-assisted draft. Hugo explicitly authorized review, push and deployment. That authorization is not comprehensive human verification. Follow-up evidence can revise the assessment.

Confidence: Medium for the directly inspected public statements and their inconsistencies. Low for practical agency effects, implementation and clinical performance.

Prepared by: OpenAI Codex / GPT-6, using the CAIHL-derived HugoScore framework. Review date is September 17, 2026 in America/Los_Angeles, corresponding to September 18 UTC. The vendor-submission email was checked against Gmail. Public findings rely on the live rendered homepage, About, privacy, terms, pricing and photo-diagnosis explainer, plus a limited validation search.

Limits: No account creation, image upload, health-data transmission, purchase, generated-result testing, network inspection, security audit, clinical evaluation, legal determination or vendor correspondence. Public terms describe commitments and claimed boundaries, not verified implementation. Submission contact details and the original private note are excluded from publication.