Eight non-negotiable principles. Published, dated, versioned. This charter governs every Swenivis product and every client engagement we sign. We update it openly when our practice changes.
One-Paragraph Summary
We use AI as a tool, not a deity. Client data is never used to train any model. Critical decisions stay deterministic. AI features are disclosed, opt-outable, and have documented owners. We refuse surveillance, social-scoring, and EU-AI-Act prohibited use cases. Concerns reach a human within 5 business days at ai-ethics@swenivis.com.
We never train models on client data.
Client data — code, documents, prompts, conversations, telemetry — is never used to fine-tune, train, or evaluate any model, whether ours or a third party's. Where LLM API providers offer training-opt-out controls (Anthropic, OpenAI, Google), we have them enabled by default on all client engagements. Client data is retained only as long as the engagement requires and is deleted on completion.
Critical paths stay deterministic.
We use LLMs for what they're good at — language, classification, suggestion, summarisation — and we keep the auditable critical path deterministic. Financial logic, access control, data integrity, security decisions, and irreversible actions are never gated solely on a probabilistic model response. The LLM informs; the deterministic layer decides.
We design for AI being wrong.
Every AI-generated output reaches users with provenance, confidence indicators where meaningful, and a way to flag inaccuracies. We do not present LLM output as authoritative when it cannot be verified. Generated content in shipping products is logged with the model, prompt template, and inputs so any issue is reproducible and correctable.
We tell users when AI is involved.
Any user-facing AI feature in a Swenivis product is disclosed in-context — labelled clearly, with an explanation of what the AI does and what models are involved. We do not pass AI output off as human-authored, and we do not hide AI-assisted features behind generic "smart" branding.
Users and clients can disable AI features.
Every shipped AI feature has a documented off-switch. End users can disable AI-assisted recommendations, suggestions, or content generation; clients can disable any AI feature across their tenant. We do not gate core functionality behind AI being enabled.
Specific categorical refusals.
We do not build: biometric surveillance systems, social-scoring systems, real-time emotion recognition for hiring or workplace decisions, automated weapon-targeting systems, deepfake systems intended to deceive, dark-pattern conversion engines that exploit cognitive bias, or systems that violate the EU AI Act's prohibited-use categories. We will turn down engagements that require these.
A named human owns every AI decision.
Every AI-assisted product surface has a documented owner inside Swenivis responsible for its behaviour. Users have a clear path to raise concerns at ai-ethics@swenivis.com — every concern is triaged within 5 business days and tracked in our public AI Ethics changelog where applicable.
We disclose our toolchain.
Models we use across products: Anthropic Claude, OpenAI GPT-4 family, Google Gemini, open-weight models (Llama, Mistral) where appropriate. Frameworks: LangGraph, LangSmith for observability. We publish material changes to this stack in this Charter's changelog so clients can audit the dependencies their data passes through.
Write to ai-ethics@swenivis.com. A named human triages every email within 5 business days.
Include: the product or engagement, the specific behaviour observed, and any reproduction steps. Anonymous reports accepted.
For incidents involving real-world harm, data exposure, or imminent risk, mark the email subject[CRITICAL]— response within 4 hours, 24/7.
Critical incidents trigger our incident-response playbook (containment, notification, postmortem) regardless of contractual relationship.
Material changes to this charter are versioned and published here.
Initial public charter. 8 principles, scope: all Swenivis products and client engagements.
See also
The broader operating principles that this AI Ethics Charter sits inside.
Tell us about your project and we'll prepare a tailored proposal within 48 hours — no generic pitches, just real strategy.
Free consultation · No commitment · Response within 48 hours
