Noah AI CAIHL draft report

HugoScore CAIHL Draft Report: Noah AI

• Status: Draft for human review
• Last reviewed: 2026-08-13
• Review method: Public-source review of the Noah AI homepage, pricing page, privacy policy (last updated 2026-01-09), and terms of use (last updated 2026-01-09), plus web searches for independent coverage and company background. This second-pass review independently re-verified every load-bearing claim in the first-pass draft of 2026-08-12. No account creation, product testing, output-accuracy or citation-quality audit, vendor interview, user interview, security audit, accessibility audit, independent validation, or legal determination was performed.
• Service: Noah AI
• Vendor: TOLONGEVITY PTE. LTD. (dba Noah AI)
• URL: https://www.noah.bio/ (second official domain: https://www.noahai.co/)
• Category: Evidence and research literacy tool

1. Executive Summary

Noah AI is a vendor-hosted, credit-metered AI research agent for the life sciences. It plans multi-step research workflows, retrieves from PubMed, clinical guidelines, trial registries, patents, and financial and pharma databases (a claimed 100M+ sources), routes steps across third-party LLMs (GPT, Gemini, Claude, Grok), and outputs cited reports, evidence tables, and export-ready files.

Under the HugoScore patient agency framework, Noah is a hybrid case. It is designed and priced for biopharma professionals, investors, and clinicians. The tool's center of gravity is industry decision support, not patient agency. But unlike HCP-gated peers such as OpenEvidence, Noah is public-facing. Any patient can sign up, receive 300 free credits, and run cited evidence searches against the same professional sources. That access is agency-relevant. It is undercut by a credit paywall, a professional register, a template privacy policy with ad-network sharing, an explicit HIPAA exclusion that pushes regulated health interactions off the platform entirely, and no medical-advice or clinical-boundary language anywhere in the public terms or homepage, despite marketing to doctors, medical students, and patients.

• Agency posture: Mixed
• Agency axis position: 55 of 100
• Confidence: Low-to-medium draft, official sources plus a confirmed absence of independent coverage

2. CAIHL Question

Who does Noah serve?

Primarily pharmaceutical companies, biotech firms, healthcare investors, consultancies, and clinicians. The homepage's own testimonials are the clearest statement of alignment: a healthcare consulting partner, a pharma clinical-strategy executive, a healthcare hedge fund partner, and a pharma head of research. Patients appear in the audience list and nowhere in the testimonials. The vendor's commercial interests (subscription revenue, credit consumption) and its industry customers' interests (speed of drug-program and investment decisions) are the dominant embedded priorities. Patients are served secondarily, as one persona among many, and are affected indirectly whenever Noah-shaped analyses inform drug development, launch, coverage, or investment decisions.

CAIHL classification: Hybrid. Vendor-hosted, public-facing professional research AI. Professionally centered, patient-accessible, patient-affecting.

3. What The Service Does

• Agent mode: clarifies scope, plans a workflow, routes each step to a chosen model and tool, retrieves and synthesizes evidence into tables and key findings with citations, runs a self-described coverage check for missing or conflicting evidence, and produces a downloadable report with underlying data.
• Search mode: searches and summarizes PubMed and other authoritative medical websites, filtering for high-value sources.
• Health Search: searches trusted sites (NIH, FDA, and similar) with parallel AI exploration and concise summaries. This is the feature most likely to be used by patients.
• Database: drug outcomes, competition, and catalysts built for evaluation, clinical analysis, and due diligence, drawing on press releases, conference materials, government filings, trial registries, peer-reviewed journals, and news.
• News and Medical Conference Hub: AI-summarized industry news and conference tracking.
• Models: GPT, Gemini, Claude, and Grok, selected per task. Hallucination mitigation is claimed via model selection, citations, and a proprietary pharmaceutical database. All vendor-stated.
• Pricing: Basic free (300 signup credits, roughly 5 Agent tasks). Starter $148.80/yr (600 credits/month). Pro $499.20/yr (2,000 credits/month). Elite $1,999.20/yr (10,000 credits/month). Fees non-refundable, subscriptions auto-renew, billing via Stripe.

4. Patient-Impact Pathway

1. Direct use: patients, caregivers, and medical students can create accounts and receive cited health-evidence summaries and reports. 2. Clinician-mediated use: clinicians using Noah for clinical insight shape diagnosis, treatment reasoning, and patient communication without patient visibility into the tool. 3. Industry-mediated use: market assessments, go/no-go program evaluations, patient-journey analyses, and investment screens influence which drugs get developed, launched, priced, and covered. Those decisions reach patients with no notice, choice, or contestability at all. 4. Data pathway: any patient who types their own health situation into Noah sends it to a Singapore-hosted service, explicitly outside HIPAA, routed across four third-party frontier-model providers, under a privacy policy that never addresses query content.

5. Evidence Table

| Source | Type | Claim supported | Notes |
| --- | --- | --- | --- |
| https://www.noah.bio/ | Official homepage | Product features, modes, sources, model roster, audiences, testimonials | Audiences include biopharma professionals, investors, clinicians, medical students, and patients. No medical disclaimer found. Verified 2026-08-13. |
| https://www.noah.bio/pricing | Official pricing | Free tier (300 credits), Starter $148.80/yr, Pro $499.20/yr, Elite $1,999.20/yr, credit metering, Stripe billing | About 60 credits per deep research task. Verified 2026-08-13. |
| https://www.noah.bio/privacy-policy | Official legal | Data collected, ad-network and retargeting sharing including Remarketing with Google Analytics, dual retention language, AI-processing opt-out in account settings, human-review promise for solely automated decisions, GDPR/US-state/Australia-NZ rights sections, Singapore governing law | Last updated 2026-01-09. Generic template. States "We do not process sensitive information." Never addresses research-query content or LLM subprocessor handling. Verified 2026-08-13. |
| https://www.noah.bio/terms-of-use | Official legal | "The Services are hosted in Singapore." HIPAA/FISMA exclusion ("if your interactions would be subjected to such laws, you may not use the Services"), AS-IS disclaimer, $1,000 liability cap, London arbitration under Cayman substantive law, class-action waiver, non-refundable auto-renewing fees, 18+, Submissions IP assigned to vendor | Last updated 2026-01-09. No medical-advice or clinical-boundary clause. Cayman Islands contact address. Verified 2026-08-13. |
| https://www.noahai.co/about | Official (second domain) | Same product, same support email, no founders, team, funding, or locations named | Company LinkedIn page exists but no leadership is named on either domain. |
| Web searches, 2026-08-13 | Absence check | No peer-reviewed studies, regulatory records, independent accuracy evaluations, or credible third-party coverage found. Only tool-directory listings. | Confirms the low-confidence rating. Testimonials remain first-name-only and unverifiable. |

6. Mixed HugoScore Profile

6.1 Agency Posture (summary judgment)

Mixed. For a motivated, English-comfortable patient or caregiver, Noah can expand critical reflection: cited syntheses of PubMed, guidelines, and trial data, with downloadable underlying evidence. But the tool is designed, priced, and evaluated for industry. Its dominant real-world effect on patients is mediated through pharma, investor, and clinician decisions they cannot see or contest. Its legal architecture (HIPAA exclusion, no clinical boundaries, offshore dispute stack, ad-tech sharing) shifts risk onto the individual user.

6.2 Service Alignment

Vendor-hosted professional research AI. Hybrid, industry-aligned with a patient-accessible surface. Primary user: biopharma professionals, investors, consultants, clinicians, researchers. Hosting and control: vendor-controlled (TOLONGEVITY PTE. LTD.), Singapore-hosted. Interest alignment: vendor and industry-customer interests dominate. Patient interests are served incidentally through the same retrieval infrastructure.

6.3 Patient Visibility. Can patients tell AI is involved? Yes

AI involvement is the explicit product identity: named LLM roster, agent workflow shown step by step. For third parties affected by clinician or industry use of Noah, visibility is nonexistent, but that is scored under choice and contestability.

6.4 Choice and Refusal. Can patients meaningfully choose? Partial

Direct use is voluntary, and the free tier lowers the entry barrier (about five Agent tasks at signup, no payment required). Meaningful sustained use requires paid credits ($148.80 to $1,999.20 per year, non-refundable, auto-renewing). An AI-processing opt-out exists in account settings, with undefined scope. Patients affected by industry- or clinician-mediated use have no choice and no notice.

6.5 Correction and Challenge. Can patients correct or challenge what the AI produces? Partial

Citations on conclusions plus downloadable underlying data make claims checkable. The coverage check claims to flag conflicts and weak support. But no structured error-reporting or correction workflow is documented. Recourse is a support email. US-state privacy rights cover correction of personal data, and the policy promises human review of solely automated decisions with legal effects, but none of this covers correcting a wrong evidence synthesis. Patients affected by third-party use of Noah outputs have no challenge route at all.

6.6 Action Support. Does it help patients understand or act? Partial

Health Search (NIH and FDA sources), cited PubMed synthesis, and evidence tables can support understanding, second-opinion preparation, and better questions. Nothing in the product is designed for patient action in the Freirean sense: no visit preparation, appeal support, record interpretation, or advocacy features. The showcased workflows (go/no-go evaluation, launch messaging, due diligence) are industry actions. The patient-journey use case analyzes patients as a market object rather than serving them.

6.7 Equity Burden (text finding)

Cost: meaningful use is credit-gated, with the recommended tier at about $41.60/month, a real barrier for most patients. Register: outputs are professional-grade reports aimed at people who already read clinical literature, with no plain-language, low-literacy, or accessibility support disclosed. Age: 18+ only. Language: multilingual UI is offered, but whether reports themselves are multilingual is not disclosed. Legal burden: the terms push users whose interactions fall under HIPAA off the service entirely, and the dispute stack (individual arbitration in London under Cayman substantive law, $1,000 cap, no refunds) is effectively unusable for an individual patient.

6.8 Data Governance (structured disclosure)

• Collected: account identifiers, billing data (via Stripe), log and usage data including searches, device and location data.
• Query content: the privacy policy never specifically addresses research questions or uploaded documents, the most sensitive data a patient would provide. Not disclosed.
• Model training use of prompts: not disclosed. Queries route to GPT, Gemini, Claude, and Grok. Downstream handling by those providers is not addressed.
• Sensitive data: the policy asserts "We do not process sensitive information," in direct tension with a product whose Health Search invites health questions from patients.
• Third-party sharing: service-provider categories include Ad Networks, Retargeting Platforms, and Affiliate Marketing Programs. Remarketing with Google Analytics is named.
• Retention: "as long as the user has an account with us" for most categories, alongside a general as-long-as-necessary clause.
• Rights: access, correction, deletion, portability via support email. Account deletion and an AI-processing opt-out in settings. Regional rights sections (GDPR, US states, Australia/NZ).
• Jurisdiction: services hosted in Singapore. Privacy notice governed by Singapore law. Contact address in the Cayman Islands.
• Compliance posture: explicitly not HIPAA-tailored. Users whose interactions fall under HIPAA may not use the Services.
• IP: "Submissions" (questions, feedback) are assigned to the vendor outright. "Contributions" remain user-owned. Which bucket a research query falls into is ambiguous.

Overall: template-grade governance with ad-tech sharing, drafted for a generic website rather than a health research tool. The most patient-relevant questions are unanswered.

6.9 Clinical Boundaries. Unclear

The profile's most striking gap. Despite marketing to doctors, medical students, and patients, neither the homepage nor the terms of use contains any "not medical advice," "consult your physician," escalation, or emergency language. The only disclaimer is generic AS-IS boilerplate disclaiming accuracy of all content. The HIPAA exclusion functions as a liability shield, not a safety boundary. In-product disclaimers may exist behind the signup wall. Not researched in this pass (no account testing).

6.10 Evaluation Ownership. Vendor-defined

Quality claims (low-hallucination model selection, impact-score filtering, coverage checks) are all vendor-stated. Social proof consists of four first-name-only testimonials from consulting, pharma, and hedge-fund users. No published accuracy benchmark, peer-reviewed evaluation, independent audit, named leadership, or patient-partnered evaluation exists. This pass confirmed the absence by search: the company's public footprint is limited to its own two domains, a LinkedIn page, and tool-directory listings.

Agency Axis Position: 55

Calibrated against neighbors in the HugoScore directory:

• OpenEvidence (18): HCP-gated, clinician-only, institution-serving. Noah sits well above this. Patients can actually use it.
• Inciteful Med (84): patient-directed by design, free anonymous access, plain-language output, explicit clinical boundaries. Noah sits well below this. It is not built for patients, is credit-gated, and its flagship workflows serve industry.
• The 55 cluster: TrialScreen (55, open patient search funded by sponsor recruitment), Counsel Health (55, patient-useful access with payer-aligned economics), and Wysa (55, patient-facing but institutionally channeled). Noah fits this cluster exactly: a genuinely usable patient surface sitting on an industry-aligned business.

The axis position is a lens-orientation cue, not a quality score.

7. Key Unknowns

• Whether research queries and uploaded documents are used for model training, retained, or shared with the four LLM providers, and under what terms. (Not disclosed.)
• Whether in-product medical disclaimers, escalation prompts, or safety boundaries exist behind the signup wall. (Not researched, no account testing.)
• Actual citation quality, hallucination rate, and coverage-check effectiveness. (Vendor-claimed only.)
• Company ownership, funding, leadership, headcount, and the relationship between the Singapore entity and the Cayman Islands address.
• Whether reports can be generated in non-English languages.
• Whether patients can request correction of a factually wrong report, and what happens when they do.
• Whether general users receive different outputs, warnings, or pricing than professional users.
• Independent verification of the 100M+ sources and proprietary-database claims.
• Data breach history, security certifications (SOC 2, ISO 27001), and subprocessor list. (Not disclosed.)

8. Patient Agency Interpretation

Noah illustrates a category CAIHL should watch closely: professional-grade evidence infrastructure that patients can technically reach but that was never designed around them. On the critical-reflection arm of the Freirean praxis cycle, Noah has real potential. A caregiver who can pay for Pro-tier credits gets industry-quality cited synthesis most patients have never had access to. On strategic action, it offers little: no advocacy, appeal, visit-prep, or record workflows. On liberation through agency, it is ambivalent. Repeated use could build research capacity in a skilled user, but the pricing, register, and legal posture select for exactly the patients who already have the most agency, while its largest patient effects flow invisibly through drug-program, launch, and investment decisions the affected patients never see.

The asymmetry is the finding. The same engine is sold to industry to model the patient journey as a market, and to patients as a way to understand their own evidence. The industry side gets the testimonials, the workflow design, and the top pricing tier. The patient side gets an 18+ signup, a HIPAA exclusion, and no clinical boundary language.

9. Publication Recommendation

Publishable as an AI-assisted draft profile. Noah is squarely in scope, fills a gap in the directory (professional research agents that patients can also reach, a bridge case between OpenEvidence and Inciteful Med), and the CAIHL findings are substantive. Before any confidence upgrade: hands-on account testing (especially in-product disclaimers and the general-user experience), a citation-quality spot check, a subprocessor and training-use inquiry to the vendor, and a periodic re-search for independent coverage.

Review Provenance

• Criteria source: HugoScore patient agency framework, CAIHL-derived. Same criteria as every tool in the directory.
• Reviewer: AI-assisted public-source draft prepared in Claude (Cowork agent session). No named human reviewer.
• AI / model: Claude Fable 5 (claude-fable-5), second-pass review of 2026-08-13. The first-pass draft of 2026-08-12 was produced by a Claude Sonnet model per the project maintainer; its own provenance section misstated the model, and its self-reported provenance should be treated as unreliable.
• Human review: none. This is an automated draft evaluation and must not be presented as a comprehensively human-reviewed profile.
• Review date: 2026-08-13.
• Limitations: no account creation or product testing, no output-accuracy or citation-quality audit, no vendor, customer, or patient interviews, no security or accessibility audit, no legal determination. Company-background claims are limited to the vendor's own documents plus a search confirming the absence of independent coverage.

Change Log

• 2026-08-12: Initial AI-assisted draft evaluation created from a public submission. Model recorded as Fable 5, later corrected to a Claude Sonnet model per the project maintainer.
• 2026-08-13: Full second-pass re-review by Claude Fable 5 (claude-fable-5). Independently re-verified all site claims, ran independent-coverage and company-background searches, added the second official domain and absence-of-coverage evidence, corrected model provenance, re-anchored the axis calibration to the directory's 55 cluster, and removed an unconfirmed Team-tier claim. Posture, axis position, confidence, and publication recommendation unchanged and re-affirmed.