Responsible AI

1. Overview

We build and run AI and agentic systems for a living, so we hold ourselves to a clear standard for how those systems are built and how your data is handled inside them. This page sets out the principles we apply. They are the same principles we write about publicly and use on our own internal tooling, not marketing gloss.

2. How We Build AI

• Verification before trust. The AI that does the work never gets the final say on high-stakes output. A separate check, with fresh context, is set up to try to break the result before anything ships.
• Guardrails by architecture, not by instruction. If an action is dangerous, we remove the capability rather than merely telling the model not to use it. Irreversible actions are gated behind a human, and sensitive operations run in a safe mode by default.
• Data minimisation. We give a model the smallest set of information a task needs, not the largest, which is better for both accuracy and privacy.

3. Your Data & AI Models

• We keep client and personal data out of model training.
• We use enterprise or API tiers from model providers, with no-training settings and appropriate retention controls, rather than consumer chat tools, for any work that touches your data.
• We consider data residency and confidentiality per engagement and design accordingly.
• The sub-processors involved in a project, including model providers, are set out in the applicable order or DPA. See our DPA & sub-processors page.

4. Human Oversight & Accountability

A person, not a model, owns the outcome. Where a system makes or materially informs a decision that has a legal or similarly significant effect on someone, we design for meaningful human review, in line with the spirit of Article 22 of the UK GDPR. We would rather a system escalate to a person than guess.

5. Accuracy & Evaluation

We measure before we trust. That means building a golden set of real cases, scoring outputs against a written rubric rather than a vibe check, and treating uncertainty as failure, because a maybe in testing becomes a mistake in production. You can read more in our guides on AI governance and using customer data with LLMs.

6. Transparency

We tell you where AI is used in what we build, and we design so that important decisions can be explained and traced rather than emerging from a black box. This is not just good ethics, it is what regulation such as the EU AI Act and the FCA Consumer Duty increasingly expects.

7. Contact