Mechanical minds in charity fundraising - considering the Fundraising Regulator's guidance on AI

Some of these tools were in the realm of science fiction not too long ago, but as they have crystallised in the fundraising landscape, they have brought with them calls for practical guidance, regulatory oversight, and ethical clarity.

So, if your charity is seeking assistance from AI to raise funds, this article is for you. We have set out, in practical terms, what you should consider, before using AI tools, in light of the guidance from the Fundraising Regulator ('the Guidance').

Trustees must lead from the front

AI systems still have considerable flaws.

• Transparency - AI often makes its decisions in a 'black box' - inputs and outputs can be examined but internal logic is hidden. This results in a relatively opaque operating model, with minimal transparency which means it is difficult to create an audit trail.

• Hallucinations - Recent examples include Air Canada's chatbot inventing a policy on bereavement, litigants presenting fake case precedents to courts, and Chat GPT claiming that dinosaurs invented some primitive forms of art. Hallucinations, like artistic dinosaurs, can be amusing to read about in the abstract but the reputational risks to a charity are obvious.

• Bias - AI can also replicate racist and sexist biases and may seek to raise funds in an unethical manner. There are well documented examples of AI systems targeting vulnerable donors, such as the recently bereaved and elderly.

These all run the risk of legal action by donors, regulatory intervention by the Fundraising Regulator and the Charity Commission, or creating potentially significant reputational harm. Given these possible pitfalls, the Guidance emphasises that trustees must lead from the front.

What does leading from the front mean?

• Risk assessments - Charities should conduct a proportionate risk assessment before a new AI tool is used to raise funds. If this step is missed, there could be unintended consequences that lead to a breach of the code.

• Upskilling - Trustees should make sure that they are upskilled to a reasonable level, with a clear understanding of the AI tools used by their organisation. The extent to which upskilling will be necessary might vary depending on the tool, but knowledge of the data the AI tool will access, how it will process that data, potential risks (e.g. accuracy, bias, data leakage, reputational harm), and mitigation strategies should be foundational to any training.

• Test deployment - Before the AI system is released, trustees should consider testing the AI in a controlled environment to make sure that the outputs are as expected. Trustees should also check if there are any risks around data loss that arise from incompatibility between an AI system and the charity's IT systems.

• Accountability - AI use should be human centred, with accountability attached to its decisions. This means there should be a "human-in-the-loop", which in real terms could involve an individual performance monitoring the AI tool to prevent algorithmic bias and checking its outputs, or building in safeguards where AI generated materials are only released to the public once they have been checked by a human. This helps to create traceability in the event of an audit, because, as noted above, it is difficult, if not impossible, to determine how the AI system itself has created its outputs. It also mitigates against compliance risks, such as with respect to the Fundraising Code more generally and charity's obligations under the Equality Act if there is algorithmic bias.

• (More) transparency - It should be made obvious to donors if an AI system is being used to target them for funds. In some cases, reference to a published AI policy will be enough. There will be other cases though, like where a charity uses AI generated imagery or an AI generated personalised appeal, when trustees should consider if they need to include express disclosure to make it clear to donors that AI has been used. Charities should always be careful using case studies which must be representative of real situations and ensure that all fundraising material does not mislead a potential donor or overstate need.

• AI policy - Trustees should adopt and publish an AI policy. This is a useful tool to provide evidence of accountability and transparency. The AI policy should, as a minimum, set out how human oversight is applied, how data is handled, and explain which AI systems are being used as part of the charity's fundraising activity.

In a nutshell, a charity's board of trustees should understand the AI system they are deploying and have a clear policy in place that deals with the risk of using it.

Legal considerations - data protection, contractual, and intellectual property law

AI does not exist in a vacuum and, like all tools in the data driven world, is open to manipulation and misuse by the untrained and ill-intentioned without safeguards. In particular, the Guidance warns of risks in the context of intellectual property ownership, data loss, and the misuse of personal information.

AI which is free at the point of use

As the saying goes, there is no such thing as a free lunch.

If a charity is making use of a publicly available system, such as a free version of Chat GPT, without payment and accepting the provider's terms, it is highly likely that there are few protections in place for any data that the charitable organisation is putting into the AI system. This could mean data input is pulled into the ownership of the AI system provider, or your organisation's confidential information ends up in the public domain.

The implications of this are significant. If you are inputting sensitive personal data, such as donors' financial data or beneficiaries' medical data, this could potentially put you in breach of your data protection law obligations and at risk of a penalty from the Information Commissioner (and in turn the Charity Commission). There are also risks to the value of an organisation's confidential information, as this could be disseminated or otherwise repurposed by AI without consent once it has been uploaded as the charity will not have control over downstream use.

Paid for AI

Paid for AI systems can come in all shapes and sizes. Charities may want to engage an AI provider to develop a custom AI system to assist with fundraising. Your negotiating power will vary depending on the commercial context, but the core tenet for trustees should be to try and retain control over what happens to the organisation's data and mitigate against the misuse thereof.

Third-party use of your materials in their AI systems

The Guidance rightly identifies third-party misuse of charity fundraising materials as a risk. There may be third-party actors who seek to use a charity's products without consent. This could take the form of using AI generated fundraising materials that incorporate your organisation's own materials or deepfake style misuse of charity branding. A robust system, such as authentication for use of the charity's fundraising materials, should be put in place to defend against such piracy or misuse. Beware though, in practice, there are limits to how effective contractual protections can be and we expect this to become an increasing concern over the next few years.

Conclusions

The Guidance is laser focused on pushing trustees to ensure accountability, transparency, and competence, when using AI systems to fundraise. Ultimately, the best way for trustees to do this is to ensure there is robust, human centred, governance in place to manage the deployment of AI tools for fundraising - and making sure to leave a clear audit trail behind.

DAC Beachcroft advises charities across all sectors on AI governance, data protection, fundraising compliance and digital transformation, as well as all aspects of charity governance. If you would like support in reviewing your AI approach or understanding how the guidance affects your organisation, we are here to help.