Responsible AI use policy

The European AI & Society Fund has a special responsibility when using AI in our work. We are working towards AI that is safe and fair for everyone. But we know we are not there yet.

Many of the AI tools that are currently available to us cause harm to people, society and the planet – including the exploitation of the workers who label data, the concentration of corporate power which undermines democracy and equity, and the environmental impact of data centres.

We need to weigh carefully where and how we adopt AI and consider where the benefit of using tools that can make our work easier and better will outweigh those costs. In doing so we think about the expectations of our grantees and funders as well as our responsibilities to our hosting organisation.

Aligning with our mission, being open with our grantees and supporting philanthropy

Our grantees have high standards. It’s important that we are transparent with them about what when we are prepared to use AI – and when we aren’t – especially when it comes to the information that they share with us. And we need to feel confident that the way we work with AI is consistent with our shared mission.

How to use AI is a live discussion in philanthropy. Our mission is about shaping AI to serve society, shifting its trajectory towards the public interest. But many foundations ask us how they should be using AI in their day-to-day operations. We want to share with them our own approach to AI adoption, recognising that other organisations with different contexts, constraints and values may reach different conclusions. We hope to point them towards values-aligned groups that can offer deeper practical expert advice to help them find answers.

Finally, the Fund is hosted by the Network of European Foundations. NEF is responsible for procuring the software systems we use and has its own AI policy that ensures our work is compliant with the law. For instance, this means we would never enter sensitive data about individuals into an AI system. Additionally, NEF’s AI community of practice has developed a guide to help teams use AI responsibly.

Our approach to responsible AI use

Building on this strong foundation, the European AI & Society Fund team has formulated its own approach towards AI use, appropriate to our specific needs as the leading philanthropic actor on AI in Europe. As both technologies develop and our own practice matures, we expect to continually update this thinking.

Our approach is guided by three questions:
1. How to keep our communities’ information safe?
2. How to preserve our expertise?
3. How to be proportionate to our purpose?

1. Safety of our communities’ information

Our grantees work on highly contested issues and face increasing pressure from both companies and governments. The information they share with us can be sensitive. For instance, identifying the individuals involved and revealing tactics and ideas which could be weaponised against them. Entering this information into a generative AI tool risks leaking it publicly. This means we go beyond the legal requirements to not process prohibited data and think about all the potential implications of AI use for the people we work with. Concretely this means us making sure we anonymise information we put in genAI tools, have clear limits about information that we never put in AI tools (in particular any confidential, personal and/or non-publicly available data), as well as consider levels of security of different systems.

2. Preserving our expertise

The European AI & Society Fund serves a unique function in sharing our insights with our networks in civil society and philanthropy and helping them work strategically and effectively together. That means it’s critical to protect our team’s thinking and judgement.

We see value in using AI where it can effectively substitute mundane tasks such as formatting, scheduling, or summaries. And we may use it where we feel it can complement our own capacities (research, brainstorming, working through a problem). But we are watchful for overreliance on generative AI that can erode skills and critical thinking over time, so-called cognitive decay. We check AI-assisted or generated content not only for factual accuracy but also for framing and overall tone. And final decisions on what to fund, what to publish, and what position to take are always made by people.

We value opportunities to learn. It can be tempting to use AI as a shortcut to distil patterns from large amounts of data, such as grantee reporting or application forms. But doing that work ourselves – reading through documents to formulate insights – builds our understanding of the field in a way that outsourcing to a model does not. For this reason, we do not use AI in our assessment of applications and reports.

3. Being proportionate to our purpose

Before we use AI, we consider whether it is the appropriate tool, not simply the most convenient one. We draw here on our grantee AlgorithmWatch’s guidelines for the responsible use of generative AI. For instance, we prefer more environmentally friendly tools (for example, a simple web search) where the outcome would be similar to a generative AI query. In practice, this becomes a matter of personal judgment for each task, but one where we will continue to discuss as a team whether we are striking the right balance.

How we currently do – and don’t – use AI

Some examples of where we currently find AI valuable in accordance with these principles:

Accessibility

Simplifying language or supporting translation, helping reach
audiences who would otherwise face barriers to accessing content.

Grammar and language checks

Reviewing text (spotting typos, correcting inconsistencies in style or phrasing) though only where data is not sensitive.

Transcribing interviews or calls, with consent

AI transcription can save significant time. We only use this solution when the interviewee or participant has been explicitly asked and has given consent. We would never exclude a participant that does not agree to transcription.

Structured research, particularly comparison

Synthesising and crossreferencing information across sources, for example in a procurement
process (while the final evaluation and decision remain with the team), always
ensuring we anonymise information and preserve confidentiality and the
safety of information.

And some things we don’t do:

• We do not use AI in our grantmaking and due diligence analysis and
decisions.
• Applicant and grantee data are never entered into a generative AI tool. There
is no pre-screening or analysis by such tools in our processes and we do not
use AI to summarise proposals.
• We never enter sensitive data about the organisations we support, the
consultants we work with, or our funding partners into generative AI tools.
• We do not use or share AI-generated content without human review and
editing.
• We do not use AI to transcribe calls unless consent has been explicitly
requested for a specific reason.