In Privacy-Preserving AI, Andrew Trask—Founder of OpenMined and Senior Research Scientist at DeepMind—breaks down one of the most pressing challenges in artificial intelligence: how to enable data-driven innovation while protecting individual privacy. Speaking with clarity and enthusiasm, he introduces the principles and tools of privacy-preserving machine learning, including federated learning, differential privacy, and encrypted computation.

Andrew begins by framing the problem: data fuels AI progress, but traditional approaches require centralizing vast amounts of personal and sensitive information. This model, he argues, is unsustainable in a world increasingly aware of privacy risks and regulatory constraints. The solution? Train AI models without moving or exposing the data.

Using clear analogies and visual examples, Andrew shows how OpenMined’s open-source technologies allow organizations—from hospitals to financial institutions—to collaborate securely by sharing insights, not raw data. He highlights real-world use cases where these tools have enabled progress in healthcare research, fraud detection, and humanitarian work.

Throughout the talk, he underscores that privacy and innovation are not opposing forces but complementary. By decentralizing AI development and democratizing access to secure computing tools, society can unlock unprecedented progress while preserving trust.

Andrew closes by inviting technologists, policymakers, and educators to join the OpenMined movement, emphasizing that the future of AI must be both ethical and inclusive—powered by collaboration, transparency, and respect for individual rights.

In this full-length OpenMined Talk, Andrew Trask—Founder and Leader of OpenMined and Senior Research Scientist at DeepMind—offers a comprehensive exploration of how privacy-preserving AI can redefine global data collaboration. Speaking with contagious enthusiasm and technical clarity, he describes the technologies that allow machine learning to thrive without centralizing or exposing sensitive data.

Andrew begins by outlining the core motivation behind OpenMined: democratizing access to secure AI tools. He explains how federated learning, differential privacy, and encrypted computation can enable organizations to learn from data distributed across the world—while keeping that data safe in its original locations. His examples include healthcare networks discovering new treatments, financial institutions detecting fraud collaboratively, and governments improving public services without risking citizens’ privacy.

He then shifts to the ethical and societal implications. Andrew discusses the need for trustworthy AI ecosystems, where privacy isn’t an afterthought but a prerequisite for innovation. He makes the case that privacy-preserving AI doesn’t just protect individuals—it enables collaboration at scale, bridging competitive and regulatory divides between organizations.

Midway through, Andrew shares OpenMined’s community-driven progress: more than 12,000 contributors globally building open-source frameworks that make privacy-preserving machine learning accessible to everyone. He highlights partnerships with universities, nonprofits, and international organizations working to deploy these tools for social good.

The session closes with a forward-looking perspective: as AI reshapes every sector, privacy will determine who participates in that future. By making privacy technology open, usable, and universal, Trask envisions a world where innovation and ethics advance together.