Rogue AI might kill everyone on Earth (including Adam Sandler). We’ve gathered the best resources for you to read, watch, and listen to, in order to form an opinion on this subject.
We've read hundreds of articles on how AI could kill everyone. Our favorite is from the New York Magazine—it is well-researched and interviewed multiple experts for quotes. Sadly it’s paywalled, so we also have a backup pick that focuses on the science.
Published Sep. 17, 2026
The best explainer so far is Dwarkesh Patel's essay "The Rise and Fall of Agent Civilizations". He also has a straight-to-camera version for those who prefer video.
Published Sep. 17, 2026
A guide to the policy write-ups that get specific about enforcement—not a manifesto, a shortlist of the plans worth reading.
Published Sep. 17, 2026
Where to start on the research, and how to tell whether the field is making progress, without having to become a researcher first.
Published Sep. 17, 2026
We hired a professional YouTube scriptwriter who's watched dozens of videos on the dangers from AI and picked the best. His top pick is a 42-minute video essay by SciencePetr, and his best shorter video is a 17-min explainer by Hank Green.
Published Sep. 17, 2026
“People have been frequently asking for [the best thing to read about AI risk]. Seems great for that to exist.”
“I like your take on AI 2027.”
“I like it.”
Our contributors have followed this research for years, and had read most of what they surveyed before we asked.
We aim to answer the questions we see most people asking, and provide content that directly addresses those questions.
We survey researchers who work on this full time about what they think is best, and what they link to people.
When we find something better, we change our picks. These articles and lists are meant to stay up-to-date.
AGI.FYI is a recommendation and review website. Our team conducts research and consults experts to find the best content and information to help people understand the global AI situation.
The project is run by Lightcone Infrastructure. Ben Pace is project lead and lead editor. The website was designed by Joanna Bregan. The contributing authors include Justin Kuiper, Aaron Silverbook, and Michael Dickens. The copy editors include Justis Mills, Vishal Prasad, and Georgia Ray. Beat Hagenlocher worked as a site engineer. The URL was a gift from Adam Scherlis.
Lightcone Infrastructure runs LessWrong, which hosts several pieces recommended on the site, and built the website for AI 2027, which we also recommend.
AGI.FYI