Responsible Use of AI Policy
September 2026
This policy was adopted on 21 September 2026, and will next be reviewed on or before 21 September 2027
This policy has been approved by: Annette Crosse, Founder/CEO (annette@egality.health)
Purpose
Egality uses artificial intelligence (AI) tools to work more effectively, but our work involves patients, communities, and sensitive personal data, and rests on trust and genuine human engagement. This policy sets out how we use AI responsibly, the safeguards we apply to data, and the uses we deliberately avoid. It applies to all Egality employees, contractors, and associates.
How we use AI
We use AI tools to support, not replace, the judgement and expertise of our team. Our current uses are:
- Research and drafting
Compiling evidence, background research, carrying out a first level of analysis on anonymised meeting transcripts where we have the appropriate permissions to do so, and drafting written material such as award applications, statements of work, and other documents, always reviewed and verified by a person before use.
- Market intelligence and business development
Analysing sector information, identifying and research potential clients, and supporting lead generation and outreach to grow the business.
- Administration
Note-taking, summaries, scheduling, and similar internal tasks.
- Meeting recording and transcription
Recording, transcribing, and summarising meetings. Our meetings can involve patient, community, and client discussions, and this use is subject to the data and confidentiality safeguards set out below: we only use meeting tools that provide appropriate data protection, we obtain consent where required, and we do not permit recordings or transcripts to be used to train external AI models. Where a meeting involves sensitive personal data, we consider whether recording is appropriate at all.
Where we choose not to use AI
Egality does not use AI to generate creative content that represents people, communities, or lived experience, and we do not use it to synthesise or stand in for a community’s voice. Our work is built on co-production; genuine partnership with the people our work represents. Using AI to fabricate a person, a story, or a community voice would undermine everything we exist to do. Where we produce communications and recruitment materials, the human and community involvement in creating them is not optional, and AI does not substitute for it.
Human oversight
A person is always responsible for any output that draws on AI. AI assists our work; it does not make decisions or sign off on outputs. We check AI-generated material for accuracy, bias, and appropriateness before it is used, consistent with our evidence-based standard, any claim originating from an AI tool is verified against reliable sources before we rely on or publish it.
Disclosure
We take a considered, case-by-case approach to disclosing our use of AI, guided by a simple principle: we are transparent wherever AI use could reasonably affect a reader’s trust in, or understanding of, our work. In practice, a member of the team decides based on context, and leans towards disclosure for public and community-facing outputs, and where AI has materially shaped the content rather than merely assisted with routine drafting or admin. When in doubt, we disclose.
Data and confidentiality
Our data safeguards are the most important part of this policy as we handle patient, community, and client data. We commit to the following without exception:
- We do not enter client, patient, or community personal data into AI tools without appropriate consent and a lawful basis for doing so.
- We only use AI tools that provide appropriate data protection safeguards, and we do not use tools that would use our input to train their models where it risks exposing confidential or personal information.
- We do not permit our data, or the data entrusted to us, to be used to train external AI models.
- We handle all data used with AI tools in line with UK GDPR and our wider data protection obligations.
To keep these commitments true in practice, we confirm what each AI tool we use actually does with our data, in particular meeting-recording and transcription tools, checking where recordings are stored, who can access them, and whether the provider uses our data to train its models. We review this whenever we adopt a new tool.
For example, we use Read AI for meeting recording and transcription and have assess it again these commitments. Read AI is SOC 2 Type II audited and GDPR compliant, stores meeting data within our own account accessible only people we explicitly share with, and does not use meeting content to train its models unless we opt in. We have set our account so that our data is not used for model training, and we apply retention settings to limit stored content. Participants are made aware of recording, and we consider whether recording is appropriate for meetings involving sensitive personal data.
We also use Claude (on a Claude Team account) for research. Drafting, initial analysis, market intelligence, and administrative tasks. Under Anthropic’s commercial terms for Team accounts, our inputs and outputs are not used to train Anthropic’s models by default, and Anthropic holds SOC 2 Type II certification and offers GDPR data processing terms. We only input personal data relating to patients, communities, or clients where it has first been anonymised, so that identifiable personal data is not placed into the tool.
Relationship to our other policies
This policy works alongside our Inclusive Language and Ethical Content Guide, which governs the ethical creation and use of content, and our Responsible Marketing and PR Policy, which requires that all claims are accurate and substantiated. Where those documents refer to our use of AI, this policy sets out the detail.
Accountability and review
This policy is owned by Annette Crosse, Founder and CEO and approved by Annette Crosse, Founder and CEO.
We are aware that AI tools and the risks around them change quickly and therefore commit to reviewing this policy at least once every 12 months, and sooner if our use of AI or the available tools change materially.