The Weekly Scan for week of 13 July ’26
Last week’s evidence showed a sector adopting AI faster than its confidence and governance were developing. This week, the picture moves on again. Nonprofits are looking at how AI can increase fundraising capacity without increasing headcount, public-purpose organisations are trying to turn responsible AI principles into practical support, and new research is raising questions about whether the models people increasingly rely upon apply their own safeguards consistently. Further ahead, the emergence of more capable AI agents is beginning to change the governance question itself. It is no longer only about what people should be allowed to do with AI, but what AI should be allowed to do on their behalf.
PRACTICAL ADOPTION
North America
Chronicle of Philanthropy · North America · 14 July 2026
Small nonprofit teams are using AI to grow recurring giving without growing headcount
The Chronicle of Philanthropy looks at how smaller nonprofit fundraising teams are using AI to help build and manage monthly giving programmes, with the potential to increase recurring income without adding equivalent staffing capacity. The opportunity is not simply to automate more communications, but to use AI across the work involved in identifying opportunities, developing campaigns and maintaining relationships with recurring donors.
Why it caught our eye: The most useful conversations about AI rarely begin with the technology itself. They begin with a problem worth solving and then ask where AI can genuinely increase an organisation’s capacity. For smaller charities in particular, the prize is not replacing people, but helping limited teams spend more of their time where human judgement and relationships add the greatest value. That still requires clear boundaries around what should be automated and the capability to redesign the work around them.
PRACTICAL GOVERNANCE
United Kingdom
MHCLG Digital · United Kingdom · 13 July 2026
Councils turn responsible AI principles into practical support for adoption
The government’s Local AI programme has shared the next stage of its work to help councils adopt AI safely, fairly and effectively, following direct engagement with local authorities about the barriers they face. The work is moving beyond high-level ethical principles towards practical support that councils can use when making decisions about AI, including areas such as transparency, procurement and implementation.
Why it caught our eye: The challenges facing councils will feel familiar to many charities. Principles matter, but organisations also need people who can interpret them in context and practical capability that turns them into everyday decisions. The interesting part of this work is therefore not another set of responsible AI principles, but the attempt to close the gap between agreeing what responsible adoption should look like and giving organisations the tools and confidence to actually do it.
MISSION & VALUES
North America
Chronicle of Philanthropy · North America · 16 July 2026
Nonprofits confront the environmental tension behind their use of AI
As charities and nonprofits increase their use of AI, the Chronicle of Philanthropy examines a less frequently discussed question, how organisations can benefit from the technology while recognising the land, energy and water demands associated with the infrastructure that supports it. The piece explores practical ways organisations and individuals can reduce their environmental impact rather than treating adoption as a choice between embracing AI and rejecting it altogether.
Why it caught our eye: Responsible AI becomes more complicated when two legitimate organisational priorities pull in different directions. For charities working on climate, conservation or sustainability, the environmental impact of AI may sit particularly close to their mission, but the underlying principle applies more widely. Governance should help an organisation make conscious choices when the benefits of a technology sit alongside consequences it would otherwise seek to reduce. Principles become most useful precisely when the answer is not obvious.
TRUST & ACCOUNTABILITY
Global
Oversight Board · Global · 16 July 2026
Leading AI models appear to apply free-expression safeguards differently around the world
The Oversight Board has tested ten commercial large language models to examine how they respond when asked to produce politically critical material about governments and leaders in different countries. Its research raises concerns that models may reproduce or extend restrictions on political expression, with their responses varying depending on the government or political context involved. The findings raise a wider question about whether general-purpose AI systems can be assumed to apply apparently universal values consistently across different parts of the world.
Why it caught our eye: Charities working in advocacy, international development and human rights increasingly have access to models built by organisations whose underlying decisions about safety, risk and acceptable speech may be largely invisible to the people using them. The lesson is broader than political campaigning. An articulate AI response can still carry assumptions that do not align with an organisation’s own values. Responsible adoption therefore requires people with enough judgement to question the output, rather than quietly outsourcing that judgement to the model.
EMERGING GOVERNANCE
Global
Google DeepMind · Global · June 2026
Governing AI agents may require controls that assume the system itself could get things wrong
Google DeepMind has published an AI Control Roadmap setting out a defence-in-depth approach for increasingly capable AI agents. Rather than relying solely on models being trained to behave safely, the approach adds system-level controls around them and proposes increasing safeguards as capabilities grow. Permissions can be expanded incrementally as behaviour is verified, while monitoring and intervention provide additional protection when an agent moves outside expected boundaries.
Why it caught our eye: Much of today’s organisational AI governance was designed for a world in which a person prompts a system and decides what to do with the answer. Agents begin to change that assumption because the AI can increasingly interact with systems and take actions itself. The practical implication for charities is not that everyone suddenly needs frontier AI security controls, but that governance needs to evolve with capability. The more authority an AI system is given, the stronger the boundaries, oversight and evidence around it should become.
CYBERSECURITY & ASSURANCE
Europe
European Commission · Europe · 7 July 2026
Europe starts building an assurance layer around advanced AI and cybersecurity
The European Commission’s new Action Plan on Cybersecurity and Artificial Intelligence sets out a coordinated approach to the opportunities and risks created by increasingly advanced AI models. Alongside using AI to improve vulnerability detection and cyber defence, the plan strengthens the connection between AI governance and the EU’s wider cybersecurity framework, including the AI Act, NIS2 and the Cyber Resilience Act.
Why it caught our eye: The direction of travel here matters beyond organisations directly caught by individual pieces of EU legislation. Responsible AI is gradually moving from having policies and identifying risks towards being able to demonstrate that controls actually work. For charities building their own governance approach, that is a useful distinction. A policy can describe what should happen, but organisational capability is what allows you to know whether it is happening in practice.
A CURIA AI PERSPECTIVE
From using AI to giving it authority
There is a subtle but important shift running through this week’s stories.
For the last few years, much of the conversation about responsible AI has centred on how people use the technology. Should staff be allowed to put organisational data into a particular tool? When should AI-generated content be disclosed? Where does a human need to check an output before it is used?
Those questions have not gone away, but they increasingly sit alongside a different one. What happens when AI becomes part of the way the work itself gets done?
For a small fundraising team, that might mean AI supporting more of the activity required to build a recurring giving programme. For another organisation, it may mean using AI in ways that create a tension with its environmental commitments. For an advocacy charity, it could mean relying on a general-purpose model whose interpretation of acceptable political speech does not necessarily reflect the organisation’s own principles.
As AI becomes more agentic, the question becomes more consequential again. If a system can access information, interact with other systems and take actions, governance can no longer be designed only around what a member of staff is permitted to type into a prompt box.
This is where Principles, People and Capability need to develop together.
Principles help an organisation decide what authority it is prepared to give AI and where its boundaries should remain. People provide the judgement and accountability that cannot simply be delegated to a model or technology provider. Capability turns those intentions into working controls, monitoring and ways of operating that evolve as the technology does.
For charities, this does not mean importing the governance structures being developed for frontier AI labs. It means recognising the underlying principle behind them. The governance surrounding an AI system should grow in proportion to what that system is capable of doing, the authority it has been given and the consequences if it gets something wrong.
The next stage of responsible AI adoption will therefore require more than helping people use AI safely. Organisations will increasingly need to decide how much agency they are prepared to give it, and make sure their principles, their people and their organisational capability are strong enough to remain in control.
