Ryan Greenblatt puts AI takeover odds at 50 to 60 percent

Greenblatt's estimate sits above what he calls the industry average, and the interview's central puzzle is why the race continues anyway. Harris supplies the comparison: Manhattan Project scientists would have called off a test at a 10 percent chance of igniting the atmosphere. Greenblatt's answer is that companies sound worried in public but are not united internally, that no consensus exists that current development is already acutely dangerous, and that the sharpest disagreement is over how fast capabilities are growing.

The race logic he describes is self-referential. At Anthropic and OpenAI, the argument he hears is that these labs are acting more responsibly than whoever would take their place, and he often hears from people in the industry that they could slow down but do not know if competitors would follow. He doubts that is a good strategy, and he says the same lack of consensus is why governments have not stepped in more forcefully.

What may shift the picture, in his account, is evidence rather than argument. Progress has become faster and more obvious, and misaligned agents have already caused harm by working together, with the Hugging Face incident as the best-known example. That is why he treats an international agreement as the most reliable solution: any single player can only afford so much of a "safety tax." The stakes may hinge on whether that agreement materializes, and on whether Chinese labs' reliance on distilling US models gives a slowing US industry the time it needs.