ONIC
Login Sign Up Download Brochure
ONIC
Login Sign Up
ONIC - One Nutrition Intelligence Centre

Ethical Use of AI in the ONIC Product Vetting & Certification System

A comprehensive guide to how AI supports - but never replaces - human decision-making within the ONIC certification ecosystem. For all stakeholders including startups, manufacturers, laboratories, and consumers.

Product Validation Claim Verification Scientific Rigor STOF (Safety, Traceability, Origin, and Finished-product verification) Validation
Section 01

Introduction

ONIC (One Nutrition Intelligence Centre) uses an AI-powered review system to support its mission of bringing trustworthy, science-backed dietary supplements to Indian consumers. This FAQ addresses how that AI system works, who built it, what safeguards exist, and how ONIC ensures that technology augments human judgment without replacing it.

This document is intended for all ONIC ecosystem stakeholders: Participating Startups, Empaneled Contract Manufacturers (ECMs), Empaneled Analytical Laboratories (EALs), Empaneled Ingredient Suppliers (EIS), regulatory observers, and consumers seeking to understand the role of AI in the products that bear the ONIC Certification Mark.

The Foundational Principle

AI within ONIC is advisory, never determinative. Every certification decision is made by qualified human assessors. The AI Engine helps these experts work faster and deeper - it does not make decisions for them. No product receives or loses ONIC certification based on AI output alone.

Section 02

The AI Engine

Understanding who built the technology, what it contains, and how it was validated.

Q1

What is the AI system used by ONIC?

ONIC's AI-powered review system is a curated AI Engine purpose-built for the nutraceutical and dietary supplement industry.

Q2

What data does the AI Engine contain?

The engine operates on a proprietary, curated database that includes:

  • Over 12.8 million scientific and commercial data points, continuously updated
  • 10,000+ ingredients mapped to targeted disease conditions and health outcomes
  • Over 1 million clinical papers, sourced from PubMed and other peer-reviewed databases, referenced to specific ingredients and dosages
  • Regulatory compliance data for approved ingredients across 11 countries (including India, USA, Japan, Mexico, UK, EU), with reference checks available for 20 countries
  • 300,000+ intellectual property and non-IP data points covering patents, trademarks, and existing product formulations
  • Toxicological profiles, heavy metal limits, contaminant thresholds, and safety interaction matrices
  • Supply chain data including 2,800+ contract manufacturers, 3,200+ ingredient suppliers, and 4,000+ packaging providers globally
  • Traditional knowledge datasets curated from Ayurvedic, Chinese, Korean, and Mesoamerican traditional texts, with over 2,700 Ayurvedic plants and their bioactives catalogued
Q3

Who curates and maintains this database?

The AI Database Library was created and is maintained by a dedicated team of 20+ subject-matter experts, including Phytochemists, clinical researchers, PharmDs (Doctors of Pharmacy), medical doctors, food scientists, regulatory affairs specialists, and data scientists. This team continuously identifies, validates, and incorporates new research data, with dataset updates published monthly.

This "curated AI" approach distinguishes the AI Engine from general-purpose large language models. Rather than generating answers from probabilistic text prediction, the engine retrieves and cross-references validated scientific data points curated by human domain experts. Every clinical reference, regulatory datapoint, and ingredient interaction in the database has been reviewed by qualified scientists before inclusion.

Q4

What AI techniques does the AI Engine employ?

ONIC AI uses a proprietary Retrieval-Augmented Generation (RAG) architecture combined with multiple AI methodologies:

  • Pathway Prediction: Artificial neural networks for predicting ingredient metabolic pathways and pharmacological interactions
  • Synergy Assessment: Machine learning models for assessing ingredient synergies and potential contraindications
  • Regulatory Compliance Mapping: AI-driven regulatory mapping that cross-references formulation components against country-specific approved ingredient lists, dosage limits, and labelling requirements
  • Clinical Evidence Analysis: NLP-based analysis of clinical trial literature to extract efficacy endpoints, study quality indicators, and evidence strength classifications
  • RAG Foundation: Validated datasets (not generative hallucination) underpin every output - the system retrieves from curated data, not from open-ended text generation
Section 03

How AI Is Used Within the ONIC System

Understanding where AI operates, what it does, and - critically - what it does not do.

Q6

What role does AI play in ONIC's product certification process?

AI operates within ONIC exclusively through ONIC Commercial Services (OCS), the commercial division. It plays no direct role in the independent certification decisions made by ONIC Certification Authority (OCA).

Function What the AI Does (OCS Side) What Humans Do (OCA Side)
Product Validation Screens ingredient safety profiles, flags contraindications, checks dosage ranges against clinical literature, identifies potential contaminant risks OCA assessors independently verify all safety claims through physical testing at EALs, document review, and expert judgment per ONIC-STD-001
Claim Verification Cross-references product health claims against indexed clinical trials, rates evidence strength, flags unsupported or exaggerated claims OCA assessors score Domain D2 (Label Transparency) and D4 (Scientific Substantiation) using the 20-point RNS rubric with human expert review
Scientific Rigor Evaluates quality of cited clinical evidence (study design, sample size, statistical significance), maps evidence hierarchy per ONIC standards Minimum two qualified human assessors review every certification decision. 100% human verification of AI-flagged rejections
STOF Assessment Models Safety, Traceability, Origin, and Finished-product verification dimensions computationally; flags gaps in traceability or sustainability documentation OCA assessors conduct full 5-domain scoring (D1-D5) including on-site audits, blind sample testing, and independent lab verification
Q7

Does the AI make any certification decisions?

No. Absolutely not.

Per ONIC's governing frameworks, all AI outputs within ONIC are advisory only, never determinative. A minimum of two qualified human assessors must review every certification decision. 100% human verification is required for any AI-flagged rejection. Every applicant has the contractual right to request a fully human review with zero AI involvement.

Q8

Why does ONIC use AI at all if humans make the final decisions?

The dietary supplement industry involves extraordinary complexity: thousands of ingredients, evolving clinical evidence across millions of published papers, regulatory requirements that differ across jurisdictions, and supply chains spanning continents. No individual human expert, however qualified, can hold all of this information simultaneously.

AI allows ONIC to screen with greater depth (checking an ingredient against the full 1 million+ clinical paper database rather than relying on an assessor's personal knowledge), greater speed (reducing pre-screening time by approximately 50%), and greater consistency (applying the same analytical framework to every product, every time). Human assessors then apply judgment, context, and experience that AI cannot replicate - including understanding of manufacturing realities, nuanced regulatory interpretation, and consumer safety intuition.

In short: AI handles the breadth; humans provide the depth of judgment.

Section 04

The Governance Structural Firewall - Keeping AI Honest

How ONIC prevents the AI Engine from compromising certification independence.

Q9

How does ONIC prevent the AI that helps formulate products from influencing the AI that helps assess them?

This is the single most important structural question in the ONIC program, and it is addressed through a mandatory Governance Structural Firewall established in ONIC's governing framework. The firewall creates two operationally distinct divisions:

  • Division 1 - OCS: OCS (ONIC Commercial Services) operates the AI Engine for formulation support, manufacturer matching, and regulatory guidance. This is the commercial side.
  • Division 2 - OCA: OCA (ONIC Certification Authority) conducts independent product certification against the ONIC Responsible Nutrition Standard. OCA is governed by an external Institutional Review Board (IRB). ONIC leadership cannot override OCA certification decisions.

The firewall mandates:

  • Separate databases, separate access credentials, and separate document management systems. Logical separation within a shared platform is not sufficient; physical or cryptographic separation is required.
  • The AI Engine used by OCS for formulation support is architecturally separate from any AI tools used by OCA for certification assessment. No model weights, training data, or inference pipelines are shared.
  • No OCS personnel may transfer to OCA (or vice versa) without a 12-month cooling-off period.
  • Annual third-party verification of AI system separation is mandatory.
  • Three firewall breaches within 12 months triggers an independent structural review.
Q10

Can OCA access the AI's formulation recommendations for a startup whose product it is certifying?

No. The Firewall Protocol explicitly prohibits OCA from accessing OCS service agreements, fee negotiations, formulation assistance records, AI Engine interaction logs, or manufacturer matching data. OCA does not know - and must not know - whether a product was formulated with AI assistance, reformulated after AI feedback, or developed entirely independently.

This ensures that every product is evaluated on its merits against the published ONIC Responsible Nutrition Standard, regardless of how it was developed.

Section 05

Data Ethics, Privacy & IP Protection

How ONIC protects startup data, ensures informed consent, and manages intellectual property.

Q11

If my startup uses the AI Engine, does ONIC keep my formulation data to train its models?

No - not without your explicit, informed, written consent. ONIC's governing framework establishes a strict opt-in-only regime:

  • Startup formulation data, engagement data, and outcomes shall NOT be used to train, fine-tune, or improve ONIC's AI models without explicit consent obtained through a standalone AI Training Data Consent form - not buried in general terms of service.
  • Startups that voluntarily provide consent receive a reduction in OCS service fees as a contractual incentive.
  • Any startup may withdraw consent at any time with written notice. OCS must cease using that startup's data for future training promptly.
  • All data used for training must be anonymized to a standard where re-identification is not reasonably possible. The anonymization methodology is auditable by the Institutional Review Board (IRB).
  • OCS maintains an AI Training Data Registry tracking consent status, anonymization dates, and withdrawal requests.
Q12

Who owns the formulations that the AI helps create?

The startup does. AI-Generated IP produced by the AI Engine during a startup's paid engagement is assigned to the startup upon delivery. OCS retains no rights to the specific formulation output. OCS's rights are limited to the underlying AI model architecture, algorithms, and general training methodology - not to any individual startup's product.

Q13

How does this comply with India's data protection law?

ONIC's data governance framework is designed to comply with the Digital Personal Data Protection Act, 2023 (DPDPA). ONIC's governing framework establishes a comprehensive Data Governance Protocol covering data classification, access controls, breach notification procedures, and cross-border transfer restrictions.

Section 06

Transparency, Accountability & Bias Prevention

How ONIC ensures the AI system is explainable, auditable, and free from systematic bias.

Q14

How does ONIC ensure AI transparency?

ONIC's AI Credibility Requirements are aligned with ISO/IEC 42001:2023 (AI Management Systems) and ISO/IEC 42006 (requirements for bodies providing audit and certification of AI management systems). The transparency framework includes:

  • Public disclosure that AI plays a role in the product assessment process
  • OCS must maintain documentation of all AI model versions, training data sources (anonymized), and material changes to model architecture
  • When the AI Engine generates a formulation recommendation, the system must produce a human-readable explanation of the key factors influencing the recommendation. "Black box" recommendations are not acceptable
  • Every formulation where the AI Engine's recommendation was a material factor must carry a disclosure notation to the startup
  • Quarterly audits comparing AI and human reviewer agreement rates, with results documented and available to the Institutional Review Board (IRB)
Q15

What happens if the AI makes an error?

ONIC's Crisis Management Plan establishes dedicated AI System Failure Response procedures covering three categories:

  • Formulation Error: AI recommends an ingredient, dose, or combination that results in an adverse event or regulatory non-compliance. Triggers crisis response: immediate product hold, independent AI audit, and regulatory notification.
  • Certification Screening Bias: AI-assisted screening systematically favors or disfavors certain product categories, ingredient types, or applicant profiles. Triggers program-wide review, re-calibration, and enhanced quarterly monitoring.
  • Data/Model Compromise: AI data breach, unauthorized model access, or AI-generated outputs leaked to unauthorized parties. Triggers data breach protocol per ONIC's governance framework.
Q16

How does ONIC prevent AI bias?

ONIC's governance framework mandates annual audits specifically designed to detect systematic bias in AI recommendations - for example, whether the engine disproportionately favors certain ingredient suppliers, product categories, or formulation approaches. The audit program includes comparison of AI screening outcomes against human assessor decisions, analysis of approval/rejection rates across product categories and applicant demographics, and review of the training data composition for representational imbalance. Audit findings are reported to the Institutional Review Board (IRB), and corrective action is mandatory where bias is identified.

Section 07

The #ResponsibleNutrition SAFE Framework and AI

How ONIC AI's founding philosophy aligns with ONIC's certification approach.

Q17

What is the SAFE framework, and how does it relate to ONIC?

ONIC AI Standards are built around the SAFE framework - Safety, Assurance, Footprint, and Evidence - which aligns directly with ONIC's five-domain Responsible Nutrition Standard (RNS):

SAFE Pillar ONIC AI Capability ONIC RNS Domain Alignment
Safety Screens ingredients against toxicological databases, heavy metal limits, contaminant thresholds, and drug-nutrient interaction matrices from 3.5M+ curated data points Safety Verification - Contaminant testing against ONIC standards, non-negotiable safety floor
Assurance Validates label claims against clinical trial evidence, flags unsupported marketing language, ensures regulatory compliance across jurisdictions Label Transparency - Accurate labelling, claim substantiation, consumer disclosure requirements
Footprint Traces ingredient origin, verifies supply chain documentation, maps sustainability indicators including sourcing certifications and environmental data STOF Integrated Validation - Sustainability assessment including sourcing, waste diversion, GHG inventory, traceability
Evidence Cross-references formulations against 1M+ indexed clinical papers, scores evidence quality (study design, sample size, statistical significance), identifies gaps Scientific Substantiation - Evidence hierarchy scoring from systematic reviews down to traditional use evidence

Note: Manufacturer Credentialing is assessed through facility audits and the ECM checklist, which operates independently of the AI Engine.

Section 08

Practical Questions for Stakeholders

Answers to the questions individual stakeholder groups are most likely to ask.

For Participating Startups

Q18

Can I opt out of AI involvement entirely?

Yes. Every applicant has the contractual right to request a fully human review with zero AI involvement. Your certification application will be processed entirely by qualified OCA assessors without any AI pre-screening or support. Note, however, that this applies to the certification process (OCA side). The OCS commercial services (formulation support, AI Engine access) are optional services you purchase separately - you are never required to use the AI Engine to apply for certification.

Q19

Will using the AI Engine improve my chances of getting certified?

No. The structural firewall ensures that OCA has no knowledge of whether you used AI formulation services or not. Your product is assessed solely against the published ONIC Responsible Nutrition Standard. A product formulated entirely without AI assistance and a product developed using every AI feature available are evaluated identically.

What the AI can do is help you identify potential weaknesses before you apply - think of it as a practice exam, not an answer key.

For ECMs and EALs

Q20

Does AI affect our facility certification or laboratory accreditation?

No. ECM facility certification is conducted through a comprehensive facility audit checklist with on-site inspections by OCA assessors. EAL technical certification is conducted through a rigorous technical assessment including blind sample competency testing. Neither process uses AI as a decision-making input.

AI's role is limited to the OCS commercial qualification track (helping match startups with manufacturers, validating formulations), which is entirely separate from OCA certification.

For Consumers

Q21

If I see the ONIC Certification Mark on a product, does that mean an AI approved it?

No. The ONIC Certification Mark means that qualified human experts at the ONIC Certification Authority have independently verified the product against a rigorous, published standard covering safety, label accuracy, scientific evidence, manufacturing quality, and sustainability.

AI may have been used as a tool to assist the commercial team in supporting the startup's product development, and it may have been used as a screening tool to help assessors work more efficiently - but the certification decision itself was made by humans, not machines. Every ONIC-certified product has been verified by at least two qualified human assessors.

Section 09

Governing Documents & Standards Alignment

The contractual and regulatory framework underpinning ONIC's ethical AI use.

Q22

Which ONIC documents govern AI use?

Document Section Coverage
Master Contract Framework Governance Structural Firewall, IP, Data Governance Structural separation of AI systems, IP ownership of AI-generated outputs, AI training data consent, data classification, breach notification
Master Document AI Credibility Requirements Advisory-only principle, human verification requirements, transparency, disclosure, bias monitoring, ISO/IEC 42001:2023/42006 alignment
Governance Framework Additional Governance Data interoperability standards, annual AI system audit requirements, Institutional Review Board (IRB) oversight of AI documentation
Crisis Management Plan AI Failure Response AI failure categories, formulation error response, certification bias detection, AI incident response protocol
Services Catalogue AI Services AI Engine access tiers, formulation support services, IP assignment upon delivery
Responsible Nutrition Standard 5-Domain RNS The human-assessed standard against which all products are certified; AI outputs feed into but do not determine domain scoring
Q23

What international standards does ONIC's AI governance align with?

  • ISO/IEC 42001:2023 - AI Management Systems (framework for responsible AI governance within organizations)
  • ISO/IEC 42006 - Requirements for bodies providing audit and certification of AI management systems
  • ISO/IEC 17065 - Requirements for bodies certifying products, processes, and services (ONIC's target accreditation standard via NABCB)
  • Digital Personal Data Protection Act, 2023 (DPDPA) - India's data protection law governing personal data processing, consent, and cross-border transfers
O
ONIC Assistant Online
Hello! I'm the ONIC Assistant. How can I help you today?