AI companion: helpful, private and controlled.

People want help that is always present, understands context and can be trusted: reminding them at the right time, answering what is needed, supporting the work at hand. But an AI that is always beside you raises hard questions: privacy — is it always listening? Dependence — will people stop thinking for themselves? Manipulation — who shapes what it says? Data control — who owns the record of the user's life?

These questions must be answered BEFORE there is a product, not after. That is why SITECH approaches the AI companion as a research direction — small, controlled trials with clear criteria — not a product rushed to launch.

• Helpful with a clear purpose: assistance serves a specific purpose the user actively chooses — no rambling suggestions, no keeping users engaged at any cost.
• Private by design: local processing (Local) where possible; minimal collection; users control their own data — pause, delete, export data.
• Transparent at all times: users always know when AI is active, what it records, and that it is AI — not pretending to be human.
• Humans keep decision-making power: users can pause, stop entirely and delete data at any time; AI does not act on its own beyond the agreed framework (Guardrails).
• No rushed commercialisation: no product claims, no published specifications, no taking orders for something not yet verified.

IEEE 7001-2021 — Transparency of autonomous systems. The standard describes measurable levels of transparency, for each audience group — users, verifiers, investigators, the public. An AI companion must be able to explain its behaviour at a level users can understand — no black box.

IEEE Std 7003-2024 — Algorithmic bias considerations. A structured framework: bias profile, identifying stakeholders, data representativeness, drift monitoring, documenting bias-related decisions throughout the lifecycle. An AI companion serves real people — bias is not a minor error.

UNESCO — Recommendation on the Ethics of Artificial Intelligence (2021). Human oversight and decision-making, privacy and data protection, the "do no harm" principle and proportionality. This is the value foundation for every AI companion design decision.

OECD — AI Principles (2019, updated 2024). Transparency and explainability, accountability, reliability and safety — three principles any AI companion must meet before talking about features.

EU AI Act — Regulation (EU) 2024/1689, Article 50. Transparency obligation: people must be told when they are interacting with AI; synthetic content must be labelled. For an AI companion, this is a minimum requirement — not an added feature.

ISO/IEC 23894:2023 — Guidance on AI risk management. Identifying, assessing and addressing risks specific to AI companions — privacy, dependence, manipulation — throughout the lifecycle.

Privacy by design: minimal collection, transparency about storage and access, users can pause, stop entirely and delete data at any time. The research direction prioritises local processing (Local) where possible, so data does not have to leave the organisation's infrastructure. The AI control layer (Guardrails) keeps AI within the agreed framework; the AI gateway (Gateway) controls who may use which AI. See how SITECH controls AI →

• Step 1 — Share your needs: what kind of assistance the organisation needs, for whom, in which context.
• Step 2 — Define the problem and constraints: purpose, limits, privacy and control requirements.
• Step 3 — Research/prototype within a controlled scope: small, supervised trials, against agreed criteria.
• Step 4 — Evaluate against transparency and privacy criteria: results measured against agreed criteria, not by feeling.
• Step 5 — Decide the next step together: continue, adjust, or stop. No product commitment.