Loading the latest edition…
Lead story
As artificial intelligence becomes more capable and autonomous, organizations are moving from broad principles to enforceable systems of responsibility. What does effective AI governance look like in practice?
Read the lead story
Original development imageAgentic AI
Real-world deployments are exposing operational risks that conventional model evaluations often miss.
Opinion
A call for coordinated action before capability, not politics, sets the terms of our future.
Director, Institute for AI Assurance
“We need assurance methods that follow a system into deployment. A model that passes a laboratory test can still fail when tools, memory and incentives change its operating environment.”
Chair of Technology Governance, Eastbridge
“Accountability must become legible. Before autonomous systems act at scale, every organization should be able to explain who approved them, who monitors them and who can stop them.”
Chief Security Officer, Meridian Labs
“Treat agent permissions as a security boundary. The most urgent work is limiting what systems can access, how long authority lasts and how exceptional actions are reviewed.”
By Nadia Clarke • July 20, 2026 • 5 min
By Tobias Reid • July 19, 2026 • 6 min
By Priya Narang • July 18, 2026 • 8 min
By Martin Ellery • July 18, 2026 • 7 min
By Samira Bell • July 17, 2026 • 9 min
By Jonas Beck • July 16, 2026 • 6 min
Expert voices wanted
We are inviting researchers, practitioners, policy professionals, security leaders, legal experts, and responsible builders to contribute clear, evidence-aware perspectives.
An independent project
AI Safety Models collects meaningful, anonymous input to help shape how we approach AI safety models.
Our methods and safeguards will evolve openly as we learn together.
Explore AI Safety Models ↗Write for AI Safety Review
We welcome informed perspectives from researchers, practitioners, executives, policymakers and people building or governing AI systems.
Join our community
Join the contributor network, suggest a topic or help us convene the people working toward a safer, more accountable AI future.
Explore the contributor network ↗Independent journalism
Help us produce rigorous reporting, durable analysis and a publication where serious disagreement can be examined clearly.
Learn how to support ↗