The Weekly Scan for week of 27 July ’26: building organisational capability for responsible AI
This week’s stories point to a shift that is becoming increasingly important for purpose-led organisations.
AI adoption is moving beyond the question of which tools organisations should use. Increasingly, the harder questions concern how AI changes the environment around them, what people need to understand to make good decisions about it, where its boundaries should sit and how much control organisations should retain over the technology they depend on.
For charities, that is already showing up in some very practical places. New research suggests that traditional brand strength does not necessarily determine whether a charity appears prominently when people use AI to find information. The DMA has published a new practical glossary intended to help people across organisations develop a shared understanding of AI. Legal-aid organisations are considering where AI can support access to justice and where caution and human judgement remain essential.
Further out, the growing debate around open-weight models raises questions about choice, dependency and organisational control, while the prospect of substantial new wealth being created by AI companies is already prompting some nonprofits to think carefully about funding, mission and independence.
Taken together, these developments suggest that responsible AI is becoming less about abstract principles or individual tools and more about the organisational capability needed to make good choices as AI becomes embedded in everyday work and society.
DIGITAL & FUNDRAISING
United Kingdom
UK Fundraising · UK · 24 July 2026
When AI becomes a new front door to your organisation
Generative AI is changing how people find information online, and new research from charity digital agency Platypus Digital suggests that some of the UK’s best-known charities are significantly more visible in AI-generated search results than others.
The research examined 15 prominent UK charities using data from Ahrefs and Semrush. Cancer Research UK, British Heart Foundation, RSPCA and Mind scored nine out of ten for AI visibility, while Macmillan Cancer Support, British Red Cross and Marie Curie scored eight.
But perhaps the more interesting finding is that traditional brand strength does not automatically translate into visibility through AI.
The report identifies practical reasons why. Research and impact information buried in PDFs can be harder for AI systems to access, while sites that rely heavily on JavaScript may also be more difficult for AI crawlers to interpret. Organisations with very broad remits can find it harder to establish clear authority around particular questions.
There are fundraising implications too. BBC Children in Need, for example, was identified as missing visibility for searches around school and nursery fundraising ideas, despite those being highly relevant areas for the organisation.
Why it caught our eye: For many charities, conversations about AI still focus largely on how employees might use tools such as ChatGPT or Copilot.
But AI is also changing how supporters, beneficiaries and other stakeholders interact with organisations from the outside.
If people increasingly ask an AI assistant rather than a search engine where to find support, which charity works in a particular area or how they can fundraise for a cause, the way an organisation structures and publishes its knowledge starts to matter differently.
This is not simply another version of SEO. It brings together content, digital infrastructure, organisational knowledge, authority and trust.
For purpose-led organisations, the important question may increasingly become not only whether people can find you online, but whether AI systems understand clearly enough who you are, what you know and where your organisation has genuine authority.
PEOPLE & CAPABILITY
United Kingdom
Data & Marketing Association · UK · 30 July 2026
A shared language for making better decisions about AI
The Data & Marketing Association has published a new Practical AI Glossary, designed to give marketers, leaders and governance professionals a clearer and more consistent way of understanding the language surrounding AI.
The glossary covers technical and operational concepts including generative AI, large language models, assistants, agents, automation, prompting and retrieval, alongside issues such as hallucination, bias amplification, prompt injection, Shadow AI and AI supply-chain risk.
It also covers organisational concepts including AI literacy, impact assessments, human oversight, transparency, explainability and automated decision-making.
The DMA makes an important point about why this matters. Terms such as agent, assistant, copilot, automation and workflow are increasingly common but are not used consistently across technology suppliers, regulators and practitioners.
Rather than adopting the terminology of any particular vendor, the glossary encourages organisations to focus on what a system can actually do, what information it can access, what actions it can take and how it is governed.
The glossary was authored by J Cromack of Curia AI and developed with review and challenge from members of the DMA AI Council and Governance Committee.
Why it caught our eye: AI literacy is sometimes treated as knowing how to use an AI tool effectively.
It needs to go further than that.
People across leadership, technology, data, fundraising, marketing, legal, governance and operations increasingly need enough shared understanding to participate meaningfully in decisions about AI.
If one person uses the word “agent” to mean a chatbot while another assumes the system can independently access information and take actions, they may believe they are discussing the same technology while actually assessing very different levels of capability and risk.
Shared language does not solve responsible AI adoption by itself, but it creates the conditions for better questions, clearer decisions and more meaningful challenge.
That is why we were pleased to contribute this resource to the DMA and make it available as a living reference as the technology and terminology continue to evolve.
RESPONSIBLE ADOPTION
North America
OpenAI Academy · North America · 29 July 2026
Legal aid puts the boundaries of responsible AI into practical focus
A discussion convened by OpenAI Academy, Everlaw and legal-sector specialists this week focused on responsible AI adoption in legal-aid organisations.
The published agenda centred on issues including confidentiality, accuracy, bias, equity, human review and client trust, alongside the practical question of where AI tools can support access-to-justice work and where caution is required.
Participants included representatives from the Legal Services Corporation, the American Arbitration Association and AI for Nonprofits Sprint.
Why it caught our eye: Legal aid is a particularly useful context in which to think about responsible AI because both sides of the adoption argument are easy to understand.
These organisations often operate under significant pressure, with limited resources and substantial demand for their services. Technology that can help people find information, support staff or reduce administrative workloads therefore has obvious potential value.
But legal services can also involve confidential information, vulnerable people and decisions where inaccurate information may have serious consequences.
Responsible adoption in environments like this is therefore unlikely to mean simply deciding whether AI is “safe” or “unsafe”.
The more useful work is identifying where it can genuinely help, what level of human oversight different uses require and which activities should remain outside the boundary altogether.
That approach is relevant far beyond legal aid. For many purpose-led organisations, confidence will come from becoming clearer about appropriate use rather than trying either to embrace AI everywhere or avoid it completely.
TECHNOLOGY & STRATEGY
International
Open Weights and American AI Leadership · US / Global · 24 July 2026
Open AI models raise a wider question about organisational control
A coalition of major technology companies and industry organisations has called on US policymakers to support open-weight AI models and avoid what it describes as premature restrictions on their development and use.
Open-weight models allow organisations to download model parameters and potentially modify and run models on their own infrastructure rather than relying solely on access through a technology provider’s hosted service.
The original coalition included organisations such as NVIDIA, Meta, Microsoft, IBM, Palantir, Hugging Face, Mozilla and the Linux Foundation. Its argument is that open models can broaden access, increase competition and give organisations greater control over their data and technical infrastructure.
Those arguments come from organisations with significant commercial and strategic interests in the future shape of the AI market, and openness does not automatically make a model safer, cheaper or more responsible.
Why it caught our eye: Most purpose-led organisations are unlikely to start running frontier AI infrastructure themselves.
But the debate matters because the AI market is still taking shape, and the choices organisations make now can influence how dependent they become on particular platforms and suppliers.
Hosted proprietary services can offer significant advantages in simplicity, capability and support. Open-weight models can potentially provide greater flexibility, portability and control.
Neither route is inherently better.
The strategic issue is understanding the trade-off.
Greater control can reduce some forms of dependency, but it can also transfer responsibility for security, maintenance, evaluation, infrastructure and model behaviour back to the organisation.
As AI becomes more embedded in organisational systems and workflows, technology choices increasingly become questions of capability and operating model rather than procurement alone.
PHILANTHROPY & PURPOSE
International
WIRED · US / International · 28 July 2026
AI wealth could create new opportunities for philanthropy, and new questions about independence
Nonprofits and philanthropic organisations are beginning to prepare for the possibility that anticipated public offerings by companies such as OpenAI and Anthropic could create a significant new generation of wealthy technology donors.
WIRED spoke with 18 nonprofits about how they were responding.
Some organisations are increasing hiring, strengthening fundraising relationships or preparing their operational infrastructure for the possibility of substantially larger donations.
GiveDirectly, for example, has raised funding specifically to prepare for a potential wave of giving, including bringing in more engineers to automate finance and HR systems and exploring the concept of a global AI wealth dividend.
Others are approaching the opportunity more cautiously.
Model Evaluation and Threat Research reportedly decided not to solicit employees of companies whose models it evaluates because of concerns that doing so could compromise its independence. Other organisations are considering whether associations with particular donors or movements could affect relationships with partners and other funders.
The amount of future giving remains highly uncertain. IPOs may be delayed, valuations may change and newly wealthy employees may ultimately choose not to donate on anything like the scale currently being discussed.
Why it caught our eye: The interesting question here is not whether an AI-driven philanthropic windfall will actually materialise.
It is what organisations do when significant new sources of funding emerge.
Purpose-led organisations understandably spend considerable time thinking about how to secure the resources required to deliver their missions. But responsible leadership also means considering where funding comes from, what expectations accompany it and whether accepting it could affect independence, priorities or stakeholder trust.
The same applies if new donors begin concentrating substantial resources around particular interpretations of problems such as AI safety, global health or effective altruism.
Funding can increase capability, but it can also influence direction.
For organisations that exist to serve a purpose, the question is therefore not simply whether money is available. It is whether the opportunity supports the mission the organisation exists to deliver.
A CURIA AI PERSPECTIVE
Responsible AI becomes real when principles turn into organisational practice
One theme connects much of this week’s Scan.
Responsible AI is moving from broad statements of intent towards the practical work of deciding how organisations will actually operate.
That distinction matters, as it is relatively straightforward to write that AI should be transparent, fair, secure or subject to human oversight. The harder task is turning those principles into decisions that people can make consistently across an organisation.
This week provides several different examples.
The DMA glossary starts with something deceptively simple. People need a shared language. Without enough common understanding of what an agent can do, what information a model can access or what human oversight actually means, it becomes difficult for colleagues to assess opportunities and risks together.
The legal-aid discussion moves the issue into real services. Responsible adoption means understanding which activities AI can appropriately support, which require stronger human oversight and where its use may simply be inappropriate.
The open-weight debate illustrates another dimension. Greater technological control can create choice and reduce some dependencies, but it also demands greater technical and organisational capability.
Even the UK charity search research shows that organisations do not control all the ways AI will affect them. AI systems are increasingly interpreting and presenting their knowledge to other people, which means choices about content, data and digital infrastructure can have consequences beyond the original purpose for which those systems were designed.
This is why we believe responsible AI ultimately comes back to Principles, People and Capability.
Principles establish the boundaries and values that should guide decisions.
People need sufficient understanding, confidence and authority to apply those principles rather than simply follow rules they do not understand.
Capability turns both into repeatable practice through the organisation’s technology, processes, governance, knowledge and ways of working.
Trust is the outcome organisations hope to preserve, but it cannot simply be declared.
It has to be earned through what an organisation actually does.
That is also the standard we try to apply at Curia AI. Our own Responsible AI Policy is public because we believe organisations advising others on responsible adoption should be transparent about how they use AI themselves.
The aim is not to eliminate every risk or pretend that every question already has an answer.
It is to create enough clarity, capability and accountability for organisations to make better decisions as the technology continues to change.
For purpose-led organisations in particular, this matters because responsible AI adoption is ultimately not about AI.
It is about protecting and strengthening the mission the technology is there to serve.
