Chapter 1
Chapter 1
AI twenty twenty-seven Update: A One Year Timeline Check April eight, twenty twenty-six Dan Schwarz By Dan Schwarz Parts of a twenty twenty-five forecast of dangerous AI have started to come true
How will the AI-twentey-seven scenario actually play out?
Start free -> Median forecast for Superhuman Coder, Apr twenty twenty-five - Jan twenty twenty-six - Apr twenty twenty-six twenty twenty-six twenty twenty-eight twenty thirty twenty thirty-two twenty thirty-four twenty thirty-six Daniel Nikola Eli
Median forecasts for Superhuman Coders, defined as "an AI system that can do any coding tasks that the best AGI company engineer does, while being much faster and cheaper" which we use as a proxy for AGI. During twenty twenty-five, AI twenty twenty-seven authors Daniel, Nikola, and Eli all drifted toward FutureSearch's conservative position. In Q one twenty twenty-six, Daniel and Eli snapped back. Sources: AI twenty twenty-seven Timelines Forecast, Q one twenty twenty-six Update.
A year ago, I sat in a room with floor-to-ceiling windows on the twelfth floor of a building in Berkeley, the center of the AI safety movement. Daniel Kokatajlo and Eli Lifland of AI Futures were explaining their forecasting models for AI twenty twenty-seven. FutureSearch forecasters (myself, Tom Liptay, Finn Hambly, Tolga Bilge, and Sergio Abriola) studied them, and produced our own adaptations of their models of the time until superhuman coders would arrive. We concluded that it would take longer than they predicted, due to R and D bottlenecks, commercial incentives at frontier labs, and government intervention.
Their scenario felt farfetched to us at the time. But one year later, many specific predictions seem scarily close to our reality, in what AI labs are doing, what their models can do, and what the US government is doing. I'm updating my forecasts.
First, this is what was published originally in AI twenty twenty-seven, in April twenty twenty-five, for when we'd have AI systems that were better than the best AGI company engineers:
FutureSearch co-authored the AI twenty twenty-seven timeline forecast showing predictions for Superhuman Coders
The original timeline forecast from April twenty twenty-five, showing each forecaster's Superhuman Coder predictions across two models. FutureSearch was the most conservative.
Over twenty twenty-five, Daniel's median slipped to twenty twenty-nine, citing "improved timelines models and slightly slower than expected progress in general." Eli moved further out to roughly twenty thirty-two. Nikola Jurkovic, SC median twenty twenty-eight, also shifted later, moving from a three-year to a four-year median. By January twenty twenty-six, Eli's median had reached "early twenty thirty-two," Daniel's had drifted to "late twenty twenty-nine," and Nikola's AGI median sat at end of year twenty twenty-nine. All three were meaningfully closer to our twenty thirty-two than where they started.
Then in April twenty twenty-six, Daniel and Eli published a major update. The METR time horizon, which they and we had greatly relied on in our model, was doubling every four months instead of their previous five point five months. Daniel was so impressed by Opus four point six that he reduced his estimate of how capable AI needs to be for the Automated Coder milestone. Eli said that the one point five year shift came primarily from "expecting faster time horizon growth" and "coding agents impressing in the real world." Both snapped back roughly one point five years earlier.
And what actually happened in the world has an unnerving number of parallels with what happens in the AI twenty twenty-seven scenario. Now, as a forecaser, it's hard to evaluate a story for accuracy, as I can't measure how many details were right or how close they were. But just read what they wrote and compare it to what happened. (If you haven't read AI twenty twenty-seven in full, I encourage you to do so now, but only if you have strong nerves, given how it ends.)
First, the scenario describes the Department of Defense beginning to contract directly with the leading AI lab:
"DoD quietly but significantly begins scaling up contracting OpenBrain directly for cyber, data analysis, and R and D, but integration is slow due to the bureaucracy and DoD procurement process."
- AI twenty twenty-seven, Early twenty twenty-six section
- AI twenty twenty-seven, Early twenty twenty-six section
In July twenty twenty-five, Anthropic signed a two hundred million dollar contract with the Pentagon. Then, in February twenty twenty-six, because of lines Anthropic refused to cross, Defense Secretary Pete Hegseth designated Anthropic a "supply chain risk," a classification normally reserved for companies connected to foreign adversaries. President Trump posted about "Leftwing nutjobs" at Anthropic and ordered federal agencies to stop using Claude.
Compare this to what the scenario predicted would happen (albeit later, in twenty twenty-seven):
"Some non-Americans, politically suspect individuals, and 'AI safety sympathizers' sidelined or fired (latter feared as potential whistleblowers)"