People first.
Planet always.
Purpose in AI.

For workforce-related projects, we will include an assessment of which tasks and roles may change.

We will seek input from affected staff and encourage clients to include training, role development and redeployment in their implementation plans. Our own approach will prioritise developing people’s skills and capacity rather than treating headcount reduction as the default measure of success.

We will be honest that automation can affect employment. We will not promise that every role will remain unchanged or claim control over decisions made by other employers.

We will encourage teams to agree how time released by automation will be used.

That might mean better customer support, professional development, more manageable workloads or time for work that has repeatedly been postponed.

We will not assume that a faster process is automatically a wellbeing improvement. Evaluation should also consider errors, checking work, workload pressure and people’s ability to switch off.

We will involve intended users in the design and evaluation of relevant services, including neurodivergent people and people with accessibility needs.

We will aim for clear language, manageable steps, appropriate user controls and alternatives where a conversational interface is unsuitable.

We will not assume that one design suits everyone or present accessibility as achieved without testing it.

Our AI tools will not be presented as substitutes for qualified mental-health care, crisis support or human relationships.

Any wellbeing-related feature will need a defined purpose, an explanation of its limits and an appropriate safeguarding assessment. Any future clinical functionality would require specialist leadership, suitable evidence and assessment of applicable regulatory requirements before launch.

We will avoid design intended to exploit vulnerability or create emotional dependency. Where people may disclose distress, we will assess appropriate signposting and escalation arrangements without promising that automated systems can reliably identify every person at risk.

For services intended for children or vulnerable adults, we will require additional age-appropriate safeguards and specialist review before deployment.

We will not use private wellbeing conversations to score employees, inform employment decisions or target advertising.

We will define what information a service needs, who can access it, how long it is retained and how it can be deleted. We will assess supplier data handling before connecting sensitive information to an AI system.

People should understand the privacy boundaries of a service before being encouraged to share personal information.

We will assess whether AI is necessary for the task and whether a simpler approach would meet the same need.

Where we use AI, we will favour appropriately sized solutions that meet our quality and safety requirements. Our design reviews will consider repeated processing, unnecessary generation, background activity and limits on automated tasks.

We will monitor overall usage as well as efficiency per completed task. Becoming more efficient should not become an excuse to ignore growth in total demand.

We will include environmental transparency in our assessment of AI and hosting providers, alongside security, reliability, accessibility and cost.

We will seek information about electricity use, emissions, cooling, water use and relevant processing locations. We will distinguish current performance from future targets, and supplier-wide claims from evidence specific to the services we use.

Where information or location controls are unavailable, we will record that limitation rather than imply that we have verified something we cannot see.

We will assess water considerations alongside energy and carbon, rather than assuming that improvement in one necessarily improves the others.

Where data is available, we will distinguish water withdrawn from water consumed and consider local scarcity. Supplier review will include questions about freshwater use, alternative water sources and pressure on local communities.

We will not describe a general environmental donation as a verified replacement for water used by our AI services.

We intend to establish an annual environmental contribution budget and review it as our use of AI grows.

Our proposed priorities include credible water-conservation, catchment-restoration and climate projects. Before announcing a partnership or claiming an outcome, we will assess the project’s purpose, evidence and reporting.

We will publish actual contributions separately from any estimates of environmental benefit. Where several organisations fund a project, we will not claim its entire impact as our own.

These contributions will sit alongside—not replace—our efforts to reduce avoidable consumption.

We will not describe our services as carbon-neutral, water-positive, zero-impact or environmentally beneficial overall without the evidence and clear scope needed to support that statement.

We will label estimates, explain important exclusions and identify the reporting period. Environmental claims should be specific, substantiated and clear about the relevant lifecycle—principles reflected in the UK Competition and Markets Authority’s Green Claims Code.

Where decisions could significantly affect someone’s rights, employment, health, safety or finances, our approach will require meaningful human oversight.

That means a reviewer with appropriate knowledge, enough information and time to assess the outcome, and the authority to challenge or stop it—not simply approve an automated recommendation.

For agents connected to business systems, we will define permissions, action limits, approval requirements and a practical way to pause operation.

We will test systems for relevant failure modes before deployment and review them after significant changes.

We will provide a clear route for concerns, record relevant incidents and assess whether a system should be corrected, restricted or withdrawn.

People interacting with our AI services should receive a clear indication that AI is involved, understand important limitations and know how to raise a concern with a person.

We will not knowingly support deceptive impersonation, exploitation of vulnerable people, discriminatory decision-making or covert surveillance.

We will challenge briefs that conflict with these principles and decline work where material risks cannot be responsibly addressed.