Main Insight
This three-part explainer series defines AI red lines as essential guardrails against irreversible harm, demonstrates their real-world implementation by governments and companies globally, and maps multiple pathways for achieving effective international coordination.
Global Red Lines for AI: A Three-Part Series
July 3, 2025
The need to get AI governance right is becoming more pressing by the day. As the world rushes to advance AI and unlock its potential, a critical issue remains underexplored: when should AI development or deployment be off limits?
Our three-part explainer series explores the concept of red lines for AI–clear, enforceable boundaries that AI systems must not cross. Vague principles, voluntary guidelines, and suggested best practices are not enough. What we need now are hard limits designed to protect the public from serious and potentially irreversible harms.
- Part 1: What Are Red Lines for AI, and Why Are They Important?
The opening piece introduces the concept of AI red lines in simple terms. It explains why establishing clear boundaries that are enforced everywhere is essential not only to prevent systemic harm but to ensure AI is developed in ways that genuinely serve the public. - Part 2: Are There Red Lines for AI in Practice Already?
The second piece examines emerging examples from multilateral bodies, regional and national governments, corporations, and civil society actors who are beginning to draw red lines regarding AI (even if they are not labeled as such). These cases show that red lines are not theoretical. They are already being put in place in response to real risks. - Part 3: How Can Global Red Lines for AI Be Established?
The final piece explores why and how red lines should be established at the global level. It outlines potential pathways and key considerations for driving meaningful progress in the years ahead.
Together, these explainer pieces aim to build a shared understanding of red lines as a practical tool for AI governance, and to help spark global momentum for drawing clear boundaries to ensure AI systems are never developed or deployed in ways that pose unacceptable risks.
Acknowledgements
We would like to thank the following people for their contributions to this publication: Stuart Russell from the Center for Human-Compatible Artificial Intelligence, Robert Trager for the Oxford Martin AI Governance Initiative, Marc Rotenberg from the Center for AI and Digital Policy, Jared Perlo from CeSIA, Saad Siddiqui from the Safe AI Forum, Karson Elmgren from RAND, and Oscar Delaney and Oliver Guest from the Institute for AI Policy and Strategy.
Additionally, we are grateful for the support of our colleagues at The Future Society: Liza Adhiambo, Sam Baskeyfield, Delfina Belli, Li June Choi, Sumin Heo, George Gor, Caio Machado, Mai Lynn Miller Nguyen, Nick Moës, and Anoush Tatevossian.
Participation does not imply endorsement of the report or its findings. The views expressed herein do not necessarily represent the perspectives of these individuals or their respective organizations. Any remaining errors are our own.
