AI Ethics Policy

Responsible AI is Non-Negotiable

EquiViral builds AI Operating Systems that are powerful, effective, and ethical. These are the principles we will never compromise on.

Our Core Principles

Transparency

We tell users when they are interacting with AI. Every EquiViral AI OS clearly identifies itself as AI-powered. We document the AI systems used in every deployment.

Privacy by Design

Data minimisation is built into every AI OS we create. We collect only what is necessary, store it securely, and delete it when no longer needed. GDPR and NDPR compliance is non-negotiable.

Human Oversight

AI OS systems at EquiViral are decision-support tools, not autonomous decision-makers. Every significant action — from lead scoring to financial reporting — can be reviewed and overridden by a human.

Bias Awareness

We actively monitor our AI models for bias in qualification scoring, recommendations, and outputs. We test across demographic and geographic variables to ensure fairness.

What We Will Never Do

We do not build AI systems designed to deceive, manipulate, or harm users.

We do not build surveillance systems or tools that could be used for illegal monitoring.

We do not deploy AI in high-stakes safety-critical applications without appropriate safeguards.

We conduct regular audits of AI outputs for accuracy, bias, and compliance.

We maintain human review processes for all AI-generated decisions that affect individuals.

We are transparent with clients about the AI models we use and their limitations.

We will not sell client data to train AI models without explicit written consent.

We report significant AI failures or unexpected behaviours to affected clients promptly.

AI Models We Use

Anthropic Claude (Sonnet & Opus)

Used for: Lead qualification, AI OS Blueprint generation, AI sales agent, automated insight generation

Anthropic's Constitutional AI approach — safety and helpfulness built into the model architecture.

OpenAI GPT-4o

Used for: Document analysis, structured data extraction, multi-modal processing for client onboarding

Used selectively for specific tasks where multi-modal capability is required.

Google Gemini Pro

Used for: Long-context document processing, multi-language support for Nigerian and UK market content

Google's responsible AI framework governs usage; data not used for model training.

Meta Llama 3 (Self-hosted)

Used for: On-premises inference for sensitive industrial clients requiring data residency

Fully self-hosted — zero data leaves client infrastructure for approved deployments.

Mistral AI

Used for: Lightweight classification, routing, and summarisation tasks inside automation workflows

European-based AI with GDPR-compliant data processing for UK and EU clients.

Cohere Embed / Command

Used for: Semantic search, knowledge base retrieval, and RAG pipelines inside AI OS deployments

Used for retrieval-augmented generation in Operations OS and Intelligence OS modules.

Custom n8n Workflows

Used for: Business process automation, workflow routing, multi-system orchestration

Open-source, self-hosted — your data never leaves our infrastructure.

PostgreSQL + pgvector

Used for: Lead data storage, conversation history, vector embeddings, analytics

Self-hosted on dedicated infrastructure. No third-party data sharing.

Google Calendar API

Used for: Meeting scheduling and calendar integration

Used only with explicit user authorisation via OAuth.

WhatsApp Business API

Used for: Automated client notifications, appointment reminders, lead follow-up sequences

Message templates approved by Meta; no unsolicited messaging.

Report an AI Concern

If you believe an EquiViral AI system has behaved unethically, caused harm, or violated these principles, please report it. We take every report seriously.

Report a Concern