1. What this policy covers
This policy sets out how our directors and associates use AI tools in our work. It covers AI chat tools, AI features built into other software, AI note-takers and transcription, and AI tools for research, writing, analysis or code. It applies to company and personal accounts used for our work.
2. Our approach
AI tools can help us work faster and think more widely. They can be wrong, and they can mishandle data. So we use them with common sense, check what they produce, and are open about how we use them.
We follow the seven principles in Harnessing AI for environmental justice, a report for Friends of the Earth that Doug Belshaw co-authored with Laura Hilliger in 2025. In short, the principles are:
- curiosity about AI, so people can make better choices
- transparency about how AI is used, and about data and algorithms
- holding tech companies and governments to account
- including diverse voices in decisions about AI
- sustainability, reducing the environmental impact of AI systems
- collaboration with communities
- an intersectional approach that keeps AI humane.
3. Choosing tools
We use a range of AI models and choose them to suit each job. We prefer open models where they do the job well. Their workings are more open to scrutiny, and they can run on infrastructure we choose, which gives us more control over where data goes. AI tools change fast, so we do not keep a fixed list. Any tool we use for work must:
- have clear, published terms on how it handles data
- not train on what we put in, or let us turn that off, which we do
- keep data secure, in line with our information security statement.
We do not use a tool that fails these tests, no matter how useful it seems.
4. What never goes into an AI tool
- Personal data about clients, participants or anyone else, such as names, contact details or health information.
- Safeguarding records or disclosures.
- Confidential material, such as anything a client has marked or described as confidential, anything covered by a non-disclosure agreement, or commercially sensitive information we have been trusted with.
- Passwords, login details or security keys.
Everyday working material is fine, such as our own notes, drafts and public documents. If a task needs AI help with something sensitive, we remove or disguise the sensitive parts first.
5. Meetings, note-takers and participant feedback
We only use AI note-takers or transcription in meetings or sessions with clients or participants when everyone present knows and agrees. Before using AI to analyse participant feedback or other project data, we remove anything that could identify people and agree the approach with the client.
6. Checking the work
A named director or associate checks every piece of AI-assisted work before it goes to a client or the public. That includes checking facts, sources and any bias in the output. AI supports our judgement; it does not replace it. We do not use AI alone to make decisions about people. Whoever puts their name to the work is responsible for it, whatever tools they used.
7. Telling clients
If AI has played a real part in something we deliver, we say so. That could be a report, written content, research or code. We say it in the work itself or in our delivery notes. We see this as good practice, not something to hide.
8. Environmental impact
AI tools use energy. We use them where they add real value, not by default, and we prefer lighter tools where they do the job. Our environmental and sustainability policy sets out our wider approach.
9. Associates and subcontractors
Associates and subcontractors follow the same rules as directors on every project. Their written agreement with us includes this duty.
10. Reviewing this policy
We review this policy at least once a year. We review it sooner if AI tools, the law or client expectations change enough to need it.