Safety researcher puts AI takeover risk at 50 to 60 percent

By Splitfeed EditorialLast updated Sep 28, 2026, 12:20 UTC

On September 28, 2026, a conversation featuring Ryan Greenblatt, chief scientist at Redwood Research, was published in which Greenblatt estimated a 50 to 60 percent chance of an AI takeover if development continues on its current trajectory. In the same exchange, Sam Harris pressed Greenblatt on that figure, pointing out that scientists working on the Manhattan Project would have halted their work at a 10 percent risk threshold. Greenblatt attributed the sustained risk to competitive dynamics specifically between Anthropic and OpenAI, and said his primary hope for changing that trajectory rests on an international agreement.

The figure is striking precisely because it comes from inside the safety research community rather than from outside critics. Redwood Research exists to make AI systems safer, and Greenblatt occupies its most senior technical role, which gives the estimate a different weight than a warning from a generalist commentator. The more substantive tension in the exchange is the one Harris raises: if the people who understand the risk most clearly are still participating in the race, what would it take for that to change? Greenblatt's answer, that every major lab genuinely believes it is the responsible actor, frames the problem as a collective action failure rather than a case of bad faith, and that framing has real implications for what kinds of interventions could work.

No reactions from the Pro-AI, Anti-AI or Middle Ground camps have been published in response to this story. The Pro-AI camp would typically argue that high risk estimates reflect worst-case thinking and that the competitive pressure Greenblatt describes is itself a driver of safety investment. The Anti-AI camp would likely treat a 50 to 60 percent figure from an insider as confirmation that voluntary self-governance has already failed and that hard regulatory limits are overdue. The Middle Ground camp would probably focus on Greenblatt's own prescription, an international agreement, as evidence that the path forward is coordinated oversight rather than either unconstrained development or a blanket halt.

The concrete thing to watch is whether any formal international process gains traction. Greenblatt framed such an agreement as the mechanism most likely to alter the current path, so any major summit, treaty negotiation or multilateral commitment in the coming months would directly test whether his conditional optimism has a realistic basis.