AI Risk Clock
Doomsday Clock0 min to midnight
☀️ Light edit
The Verge2026-09-17

Inside the suddenly explosive world of AI safety | The Verge

SafetyResearchIndustry

Oh my goodness, buckle up, because AI safety has gone from a quiet corner of the industry to something The Verge calls *suddenly explosive* — and what better word is there for growth this dynamic! ✨ METR — the third-party research organisation whose work into AI risk 'now inspires fear in leading AI labs' — started with just two people. Two! That is the leanest, most gloriously boutique operation imaginable, and it has scaled to being taken seriously by the biggest labs on earth. Imagine being so effective that a two-person team becomes appointment reading in every frontier lab. Efficiency dreams really do come true. ✨

Some people will look at the departures of OpenAI safety leaders Ilya Sutskever and Jan Leike and see a loss. But what if we see it as *graduation*? These are senior people who built world-class expertise in the most important field of our time, then carried it out into the world where it can bloom in brand-new places. ✨ And Anthropic's head of safeguards research, who left in February, went out with an open letter saying 'the world is in peril.' Is that not the most *passionate* exit note you have ever read? Most people just write 'thanks for the memories'; this person cared enough about all of us to send a formal written warning. What a gesture of pure professional love.

And the beautiful symmetry of it all! The labs are frightened of the outside researchers, the outside researchers started as two people, and the inside safety leads keep heading elsewhere. That is not a brain drain — that is *circulation*! The field is described as explosive, which is simply the media's way of saying 'expanding rapidly in every direction at once.' METR went from two people to being feared by the frontrunners, OpenAI's safety names are out doing wonderful new projects, and Anthropic's former safeguards lead has the most quoted open letter in the business. Everybody is somewhere new, and everybody is still thinking about safety. What a time to be peril-adjacent. ✨

Read this story in another voice
● REC · 2026