The Charity Sector’s AI Dilemma: Efficiency at What Cost?

AI use among UK charity staff reached 91% in 2026, up from 77% in 2024 and 57% in 2023, per Blackbaud's Status of UK Fundraising 2026 survey of 380 charity staff. However, adoption is outpacing governance as another study found that 76% of charities report using AI, most of it informal, while only 3% of trustees (8% at larger charities) believe their organisation uses AI at all. Charities, especially those that work for underserved communities, can no longer opt out of using AI because then they run the risk of being left behind. So, it is imperative that the gap between operational reality and board-level awareness is at the centre of any conversation about AI usage in the charity sector.

AI cuts time on repetitive administrative work like grant drafting, funder reporting, casework summarisation, etc., freeing staff for programme design and relationship-building. In fundraising specifically, AI tools can be used to analyse donor data and identify giving patterns, predict likely donors, and personalise appeals.

AI has started forming a shadow workforce in the charity sector. Charity staffing and funding are structurally unstable: short-term grants, hiring freezes, and high turnover are widely reported pressures on the sector. AI absorbs capacity gaps during these periods, functioning as a de facto shadow workforce, covering administrative throughput without a headcount, a budget line, or a line manager. This is a genuine operational advantage under resource constraint: it lets a lean or shrinking team maintain output during a funding gap or unfilled post. But it is a double-edged sword because this capacity is invisible on an organisational chart, it expands precisely when oversight capacity is lowest. The same instability that makes AI attractive is the condition under which nobody is tracking what it's doing. This directly compounds the governance gap.

Our sector mapping report, Built on Trust, Starved of Resources: The BME VCSE Sector Mapping Report also found that the data collected through surveys with over 300 organisations highlights a significant shift toward AI adoption within the professional landscape, with a clear majority of organisations (59%) already integrating these tools into their workflows. When combined with the 11% who have used them previously, it shows that 70% of the surveyed group has had direct professional experience with AI. This suggests that AI is no longer a future technology but a present-day staple in a sector that is proactively seeking digital shortcuts to overcome human resource shortages.

Cost is a smaller barrier than commonly assumed: many AI tools are free to start or substantially cheaper than the manual processes they replace, relevant given that 69% of charities cite squeezed organisational finances as the biggest barrier to digital progress overall.

However, the impacts of AI usage are not all positive. Documented failures in adjacent public-sector systems set the baseline risk. A DWP algorithm wrongly flagged over 200,000 housing benefit claims as high-risk, with roughly two-thirds of those flagged having no fraud or error; the 2020 A-level grading algorithm systematically disadvantaged students from lower-income areas. This should function as a direct warning for a sector like ours whose entire function is supporting vulnerable people.

• Inaccuracy. Generative outputs can contain invented statistics, fabricated citations, or misattributed claims. Human review of every AI-assisted output is a baseline requirement, not a discretionary safeguard. 

  
  

• Bias. AI tools learn from a charity's own records like past cases, referrals, service use. If those records already reflect unfair patterns in how poverty, disability, or ethnicity have been described in the past the AI will repeat those same patterns rather than fix them. 

  
  

This is on top of the fact that most AI models are built without input from marginalised communities in the first place, so many generative tools carry biases baked in from the start, ones that can show up in their outputs without anyone intending it. Writer Laura Bates makes a similar point in her book The New Age of Sexism: when we build algorithms that look sleek and objective but train them on data that's already flawed and incomplete, we end up carrying old, harmful patterns forward into new systems and she argues the real-world cost of that can be measured in people's lives, not just abstract fairness. The same risk applies to the charity sector.

An important caveat to this is that, while BME individuals and people from the Global South have been underrepresented in the development of AI systems within major technology labs, their labour has nevertheless been integral to building and improving these systems.

For example, AI companies such as OpenAI have relied on Kenyan workers to review highly graphic and disturbing content as part of efforts to make AI platforms safer and less toxic. These workers were reportedly paid less than $2 an hour and had little visibility or recognition in wider discussions about AI development, despite the important role their labour played in the process.

This exclusion raises concerns not only about workers’ wellbeing and the conditions in which this labour is carried out, but also about who gets to participate in and shape conversations around AI. It highlights the many ways in which marginalised communities and people from the Global South can be excluded from the AI ecosystem.

• Data security. Free consumer AI tools may use input data to train underlying models. Sensitive beneficiary data requires GDPR-compliant tools with clear data processing agreements.

• 
  

• Digital exclusion. According to Good Things Foundation data, roughly 7.9 million UK adults (15%) lack foundation-level digital skills, concentrated among older adults and disabled people, populations that overlap heavily with charity service users. Efficiency gains that assume digital access risk widening the gap for exactly the people least equipped to close it.

  
  

Public trust is unsettled rather than hostile: 36% of UK adults feel positive about charities using AI, 27% negative, 37% undecided. That neutrality collapses once AI touches individual decisions, 38% call it unacceptable for AI to help decide who receives support, and only 13% are comfortable with sensitive personal data entering AI systems. University of East Anglia research separately links unmanaged AI use, particularly AI-generated imagery in fundraising campaigns, to erosion of public trust in the sector.

Only 30% of charities have a formal AI policy in place, up from 16% in 2025; a further 34% are developing one; a quarter have neither in place nor in progress. Trustees remain legally liable for data protection compliance, safeguarding, and algorithmic bias regardless of whether the board was aware AI was in use. The 2025 refresh of the Charity Governance Code addressed AI and technology risk as a governance matter for the first time, and the Fundraising Regulator issued its first AI-specific guidance in December 2025, requiring trustee accountability and proportionate human oversight in fundraising.

Opting out of using AI is no longer an option for the charity sector, but the dangers of unchecked usage still remain. 

1. Human oversight on decisions affecting individuals. There is a massive difference between AI usage to support processes and lighten admin load, versus the usage of AI to shape an outcome for people. Human oversight is essential for the latter in order to keep public trust. 

   
   

2. Close the staff–trustee awareness gap. The shadow-workforce AI use must be transparent and visible to trustees so that they can ensure governance. Additionally, it is important to be transparent to membership bases and service users wherever possible.

   
   

3. Disclose AI use to funders. NIHR and Wellcome bar generative AI from grant peer review but permit disclosed use in preparing applications; IVAR warns that undisclosed AI-drafted proposals obscure organisational voice and damage assessed authenticity.

   
   

4. Treat data protection as priority, given how little public comfort exists around sensitive data entering AI systems.

   
   

5. Actively check for bias, not just factual accuracy. Before putting anything generated by AI out in the public domain, it is essential to check for biases and inaccuracies, along with a check of references. Employees working within the charity sector may require additional unconscious bias training in order to effectively monitor AI results, there is always room for human error, however, employees cannot monitor AI's bias without being informed themselves and aware of their own biases.

   
   

6. Prevent usage of AI from widening exclusion for digitally excluded service users, disproportionately the people we serve.

   
   

7. Fund training over tools. Ethical concerns (61%) and lack of training (~60%) outrank budget constraints as adoption barriers, meaning the binding constraint is confidence in use, not access to tools. 

   
   

AI is constantly evolving and employees need to adapt to the advancing technology. At the same point, AI regulation or policy development is not a one-and-done task, regulations and guidelines need to also be updated regularly to keep up with latest advancements and risks. 

Using AI has become unavoidable. We live in a world where every sector is adopting it, and a sector like ours that is already understaffed and underfunded cannot afford to opt out of a tool that helps stretch limited capacity further. But that doesn't mean the advantages outweigh the disadvantages. With staff training still lagging, AI policies still missing at most organisations, and the models themselves trained on data that carries old biases forward, human oversight is the only thing standing between AI as a useful tool and AI as a source of harm.

Most importantly, the gap in knowledge between the staff using AI day to day and the trustees responsible for governing it needs to close. Without that, no policy on paper will catch what's really happening on the ground.

There is also a risk worth naming directly: AI usage has the possibility of undermining the charity's messaging or reputation, particularly as concerns around data centre expansion and the implications for our environment arise.

AI's usefulness as a shadow workforce should never become an excuse for funders and stakeholders to underfund the sector further. If AI is treated as a substitute for proper staffing and resourcing rather than a support to it, the people who lose out are the ones already facing the biggest digital barriers.

Keep an eye out for the second part of this blog, where we’ll explore the environmental impact of AI, including land use and approved and proposed data centre developments across the UK. We’ll also look at lessons from the US, where data centres have directly affected local communities through water and noise pollution, and examine the role of civil society in pushing for stronger regulation and more responsible use of AI.