Who Owns the AI We’re Using?

Foto von Jornada Produtora auf Unsplash

by Darwish Thajudeen 

Across Europe, community groups, charities, and NGOs are already using AI to translate documents for refugees, summarise case notes so staff can spend more time with people, and analyse satellite images to anticipate floods and protect homes. These are powerful examples of “AI for good” in practice, and they reflect MI4People’s focus on using AI for public welfare, health, environment, and society (MI4People).

At the same time, they raise a quiet but crucial question: Who actually owns the AI we’re using and who decides what it can and cannot do? The easy answer is “the company that built it,” but current research on AI sovereignty suggests a more layered reality, with control spread across companies, governments, and technical infrastructure (Brookings background). When an organisation uses AI to draft a report or analyse community data, it is not simply interacting with neutral software; it is entering an ecosystem shaped by rules, incentives, and dependencies that influence which features exist, who gets access, and what counts as acceptable use.This is why deeper questions matter:

Who controls the data being entered? Who decides how the system evolves? Who sets the rules and can change them? Who can truly see how decisions are produced? And what happens if access suddenly changes or disappears? These are not just IT questions; they are questions about trust, independence, and democracy, as highlighted in recent work on human rights and AI sovereignty (Just Security). At MI4People, the view is that organisations working for people and planet need to ask them now, before AI dependence quietly hardens into invisible infrastructure that becomes difficult to question or change.

Layers of Control

To understand who really owns the AI an organisation uses, it helps to see three overlapping layers of control.

First, there is corporate ownership. The company that builds the model sets prices, decides who can use which features, writes the terms of service, and chooses the direction of future development. NGOs already know this dynamic from social media and cloud platforms, where a product decision made elsewhere can suddenly disrupt communication or workflows. AI now extends that dependency into more sensitive areas like translation, summarisation, and risk analysis, meaning the ability to serve communities can hinge on decisions taken far outside local context and priorities (Brookings).

Second, there is governmental control. Governments increasingly treat advanced AI systems as strategic infrastructure, similar to energy networks or telecoms, and are building national and regional frameworks to manage them (Brookings report). In the United States, the National Policy Framework for AI frames frontier AI as a national security issue and pushes toward stronger federal oversight (CSET). In Europe, the Tech Sovereignty Package and the EU Open Source Strategy explicitly connect AI, cloud, chips, and open technologies to digital autonomy (European Commission; EU Open Source Strategy). As a result, access to certain capabilities may depend on where an organisation operates, which sector it serves, and how policymakers define risk and sovereignty.

Finally, there is organisational sovereignty, the level where agency becomes tangible. The practical questions are straightforward: Can the organisation choose where its data lives? Can it understand how AI-supported decisions are produced? Could it move to another provider without losing its history? Can it adapt tools to local languages and contexts? Can it keep control over critical workflows? Research on sovereign AI increasingly argues that open source, open standards, and interoperability are central to answering those questions well (Linux Foundation; Linux Foundation Europe). If the honest answer to most of these is uncertainty, then the organisation may be renting intelligence rather than truly owning its digital future.

Data Sovereignty

“Data sovereignty” can sound abstract, but at its core it is about whether organisations can make informed choices about the infrastructure and data they depend on while still collaborating globally (Brookings). For civil society, that means knowing who can access information now and in the future, which laws and regulatory regimes apply, how data is processed and combined, and how reliance on particular tools is managed over time (EU Open Source Strategy). This matters because NGOs often hold highly sensitive material: testimonies from vulnerable groups, health records, migration stories, education data, human rights documentation, and community knowledge built over years of trust (MI4People). Safeguarding that trust cannot be reduced to “our provider is compliant”; it requires technology choices that align with organisational values and the expectations of the communities being served, a theme echoed in civil society AI governance priorities (The Future Society). Put simply, data sovereignty asks whether AI choices are strengthening that trust, or quietly weakening it.

Openness, Dependence, and Accountability

Public debates often frame AI as “open source versus Big Tech,” but for NGOs this is not the most useful question. Commercial providers have built powerful tools that can improve how organisations work, while open source projects and open standards offer transparency, flexibility, and independence that proprietary systems rarely match (Europe’s digital sovereignty starts with open source; Linux Foundation). The more practical question for civil society is whether it is becoming dependent on systems it cannot understand, inspect, or replace. Many organisations have already experienced this with social media and cloud platforms, where strategies built entirely on external services became fragile when algorithms, terms, or business models changed (FSFE). AI creates an opportunity to learn from that experience: commercial tools can still be used where they make sense, but open and inspectable alternatives should be deliberately strengthened for core workflows and sensitive data.

At the same time, openness is not automatically ethical. An open model can still be biased, and an open platform can still be misused. What openness does offer is the possibility of real accountability (Linux Foundation). When code, documentation, and governance processes are visible, even partially, independent researchers can examine them, communities can raise concerns, and organisations can adapt systems to better fit local realities, with problems discussed in public rather than hidden inside proprietary logic (LFAI & Data). For NGOs, the key question is not simply whether a system is labeled open source, but whether enough about the technology can be understood to trust it with the mission.

Civil Society’s Role

Most high-level AI policies and products are currently shaped by governments and large companies, but civil society is not merely a spectator (Brookings initiative). Consultations and surveys show that NGOs and community organisations bring crucial perspectives: the lived experience of vulnerable people, local languages and cultures, the social context behind data, and a clear sense of what genuine impact looks like on the ground (The Future Society; AI governance priorities). AI should not be designed only for markets; it should also be designed for societies. Civil society helps make that happen by asking for transparency before adopting tools, insisting that governance frameworks include affected communities, and choosing open, participatory approaches wherever possible (Data & Society). Every technology decision sends a signal. Each question about data control, explainability, or how easily a provider can be changed helps push the wider ecosystem toward greater accountability (Partnership on AI).

Why MI4People Chooses Openness

For MI4People, openness is not a detail, it is part of the organisation’s identity as an AI-for-good initiative working with partners across Germany and beyond (MI4People; Pro Bono Allianz). The work spans public welfare, health, environment, and society, and rests on the belief that organisations should understand the technologies they depend on and have a voice in how those technologies evolve. That is why the tools and approaches prioritised are open, transparent, privacy-conscious, explainable, and interoperable, consistent with broader ethical frameworks for a good AI society (AI4People / PubMed).

Open standards and open tools are not simply engineering preferences. They are pieces of democratic infrastructure: they help communities, researchers, public institutions, and NGOs collaborate; they keep organisations independent enough to pursue their missions; and they allow accountability to be woven into digital systems from the start (EDRi; Linux Foundation). Openness keeps people active participants in the future of AI, rather than passive consumers of decisions made elsewhere (AI Now Institute).

When asking who owns the AI being used, it becomes clear that ownership is shared and contested across companies, states, and infrastructures (Brookings). A more constructive question is: who should AI work for? The answer here is simple: people first, always (MI4People). The future of AI should be open enough to inspect, transparent enough to trust, and accessible enough to benefit everyone (EU tech sovereignty communication; Linux Foundation Europe). Technology built for public good should never become something the public cannot understand, influence, or challenge.

The future of AI should not be written behind closed doors. It should be co-created with communities, for communities, and increasingly by communities themselves (NGI Commons; AI for Good). As organisations consider the tools they already use and those they may adopt next, one human question stands out: How can AI choices honour the people being served and strengthen, rather than weaken, their ability to shape their own lives? MI4People’s commitment is to explore that answer together and help build open, transparent, and accountable AI that truly serves the public good.