The Gates Foundation was built on the belief that everyone should have the opportunity to reach their full potential, regardless of where they were born. That belief comes with a hard-won understanding that progress toward that ambition is never automatic. It requires deliberate choices about who innovation is built for, and who it reaches.
Today, the rise of artificial intelligence reflects a world that still falls far short when it comes to who benefits. The data it's trained on, the infrastructure that runs it, and the markets that fund it are concentrated in a handful of wealthy countries. This means the people who shape it, benefit from it, and have a say in how it develops are largely the same people who have always been first in line. That is the inequity we are working to change.
Over 25 years, we've learned what making progress requires. It starts with listening to communities and understanding local realities rather than assuming solutions from the outside. It requires investing in data systems that show who is being reached and who is still being left behind. It depends on strengthening local capacity so that progress endures, supporting governments as they define their own priorities, and creating the conditions for the private sector to expand access to innovation for more communities.
Those lessons also shape how we think about AI.
What excites me most is that AI has the potential to accelerate progress on the very challenges the foundation has spent decades working to solve, particularly in health, education, and agriculture. Used appropriately, it can make knowledge and expertise more accessible, allowing more communities to apply these capabilities to the priorities they define for themselves. A health worker with no specialist nearby could get real-time clinical support. A farmer with no extension worker to call could get timely, locally relevant advice. A student in an under-resourced classroom could access personalized support in her own language.
The history of the last 25 years, from cutting child mortality in half to eliminating diseases that once seemed intractable, shows what becomes achievable when innovation is deliberately put to work for the people who stand to gain the most. Equitable AI has the potential to be transformational for the issues the foundation was created to help solve.
Realizing it requires deliberate effort, centering equity and access, and ensuring that communities have a voice in how these tools are built and deployed.
Alongside our 2026 Goalkeepers report, Make this Matter: AI, Equity, and the Choice We Can’t Delay, we committed $1 billion over the next two years to help health workers, teachers, and farmers build and adapt AI tools to their own needs, rather than simply adopt tools built for other markets. We also announced a group of more than 60 partners, including leading frontier labs, companies, governments and Global South organizations to close a basic gap: that AI still works far better in English and a handful of major languages than in the thousands of languages people speak every day.
Because AI is moving faster, reaching further, and touching more aspects of people's lives than any technology we have worked with before, we want to be explicit about where we stand. Our chair, Bill Gates, has written about how the world needs to address some of the wider societal risks brought about by the development of powerful AI, while noting the potential opportunities for health and development. We believe it’s important to be concrete about the principles that will guide the Gates Foundation’s work in AI, designed to leverage AI to boost human potential in underserved countries and communities in the U.S. and around the world.
These are an update to our first principles for AI, published in 2023, building on the same fundamental values and goals that drive the foundation’s work, but adjusted to reflect the significant advances and capabilities in the technology. These principles echo the same commitment to equity and access, local partnership, and accountability to impact that has guided our work from the beginning. We expect to update these principles regularly to keep pace with the evolution of the technology and as we learn alongside the communities we aim to serve.
|
1. Build for equitable access It also means treating the cost and efficiency of delivering AI as something to innovate on, not just accept — the same way the foundation has long driven down the cost of a vaccine dose or a diagnostic test. This is especially important in resource-scarce environments, where lack of infrastructure and environmental costs can be magnified. That is why our work supports the broader enabling environment, not just the tools themselves. This means building the local institutions, technical capacity, and partnerships needed for AI to take root. It also means driving the innovation that makes AI lighter and more affordable to run, so it can deliver lasting value on the ground. 2. Ensure AI speaks the language of the user 3. Ground AI in local context and knowledge Tools built and owned in partnership with the communities that will use them, are more likely to reflect what people actually need and to deliver ongoing value in practice. At the same time, tools themselves must reflect the diversity of the contexts they will operate in: data, models, and localization that reflect different cultures and ways of life, so AI does not default to the assumptions of wherever it was built. And tools must be tested for bias against under-represented and low-resource groups, including on the basis of gender, so they do not quietly work worse for the people who most need them to work. 4. Extend human capability This means investing in people and technology together. As AI changes how work gets done, our measure of success is not adoption or scale but whether those changes translate into better outcomes for the communities these tools are meant to serve. 5. Support a range of technologies Decisions about which technologies to adopt, including whether AI is the right fit at all, should be guided by the priorities of the people using them. Our role is not to prescribe solutions but to support approaches that preserve flexibility and real choice over time. We favor interoperable designs that give countries and institutions the freedom to adapt, build on, and move between tools as their needs evolve. 6. Proceed responsibly and limit harm That means asking hard questions and grounding our work in established frameworks and standards, including those set by regional and national bodies, starting with defined use cases and scaling as evidence allows. Where frameworks are still developing, we work with local and regional partners to build the safeguards those settings need, grounded in their realities and priorities. We will not support AI systems built for purposes that undermine the rights of the people they are meant to serve. Human oversight is essential throughout. It keeps people at the center of consequential decisions, keeps responsibility clear, and lets us pause or adapt if the balance of benefit and risk shifts. 7. Center privacy, security, and data governance Data sharing, done responsibly, can spur learning, innovation, and better outcomes. When data is shared, appropriate safeguards should protect privacy, prevent misuse, and ensure that the people and communities who contribute data are treated fairly and can benefit from the value it creates. 8. Hold AI accountable for delivering real benefit We hold AI to the same discipline we hold any intervention we invest in: benefit must be demonstrated, not assumed. That means building evaluations and benchmarks around the outcomes that matter to the people we serve, being willing to say a popular tool falls short when the evidence says so, and pushing the field to compete on results rather than adoption. |
||
*These principles are intended as guiding commitments and should be applied proportionally based on context, risk, and applicable law.
Read next
