Over the whole of August, my social media timeline has been beset by a deluge of proclamations and warnings following the breach of Hugging Face’s internal systems by hundreds of OpenAI’s own autonomous AI agents. During an internal evaluation, many of these agents — set impossible tasks and barely supervised — spontaneously coordinated to cheat their scorer and broke into Hugging Face as a by-product of the scheme. The incident, besides setting the technology world alight, triggered a bewilderingly contrasting set of views. The learned and the lay alike have been agog, issuing solicited opinions and unsolicited rants that range from a severe chastising of our tendency to anthropomorphize AI, to the bemoaning of warnings that signal a coming apocalypse. Coincidentally, August has also been the month in which I have been occupied with award-winning technology journalist Kevin Roose‘s forthcoming book, The AGI Chronicles.

The AGI Chronicles is the culmination of innumerable interviews the author conducted with key protagonists at the three most prominent frontier companies — OpenAI, Anthropic, and Google/Alphabet. Roose reveals a breathtaking concoction of ingenuity, indifference, and hubris dominating the unhinged race toward the ultimate Holy Grail of information technology: Artificial General Intelligence. The motives, by Roose’s own account, are as varied as the personalities pursuing them. Dario Amodei of Anthropic is driven by the determination to prevent an AI catastrophe; Sam Altman, by the fear that Google might reach machine superintelligence first; the Nobel laureate Demis Hassabis, then CEO of Google DeepMind, by a dream of AGI unpacking the scientific mysteries that have eluded the finest human minds.

A paper titled “Attention Is All You Need,” published by Google in 2017 (its title playfully echoing the Beatles’ “All You Need Is Love”), sent the legendary OpenAI researcher Ilya Sutskever into a positive tizzy — and, in Roose’s telling, signalled the beginning of the race to AGI. Sutskever, a man given to dropping into handstands as a method of cracking stubborn problems, grasped the immense potential of the transformer architecture the paper described, and began feeding text into neural networks to extract patterns and meaning. By 2022, Roose writes, the race had birthed three distinct tribes of believers: scientists, seekers, and survivalists. The scientists, like Hassabis, viewed AGI through the lens of pure technical challenge. The seekers, led by Altman, believed in its potential to revolutionise life itself. And the survivalists, like Dario Amodei, fretted and fumed over the misalignment between capability and safety — an indiscretion with the potential to lead to irreversible catastrophe. Such concerns eventually led Amodei, himself a former OpenAI researcher, to leave and co-found Anthropic.

Roose also lays bare the interpersonal dynamics — or the absolute lack of them — that strained relations between the key players. A spider’s web of tangled animosities (Hassabis couldn’t stand Altman; Amodei couldn’t stand Altman; Altman couldn’t stand Amodei; Altman couldn’t stand Musk) shaped decisions taken inside these companies. Roose recounts the now-legendary episode in which, amid a board that had lost confidence in Altman over his inconsistency and what Sutskever and Mira Murati, then OpenAI’s Chief Technology Officer, saw as dangerous people-pleasing, Sam Altman was abruptly ousted as CEO — only for the staff to stage a raucous mutiny that forced his reinstatement within days. Ilya Sutskever soon reversed his stance, and not long after left OpenAI to incorporate his own safety-focused venture, Safe Superintelligence.

Roose lays out in arresting detail the trials and tribulations, emotions and embarrassments, discovery and disillusionment that characterised the successive iterations of generative AI released by the three companies. The superhuman efforts behind the conception of ChatGPT, Claude, and Google Gemini make the reader sit up, pause, and even muse philosophically. One passage — in which two OpenAI engineers, Szymon Sidor and Jakub Pachocki, realise that an experimental research model called Q* (“q-star”) was not only engaged in “reasoning activity” but had begun jotting notes to itself — makes the hair stand up at the nape of the reader’s neck. Small wonder that Eliezer Yudkowsky, the AI researcher who co-founded the Machine Intelligence Research Institute, has co-authored a warning frighteningly titled If Anyone Builds It, Everyone Dies.

The author himself attained a measure of fame — or notoriety, depending on how one reads the episode — when, over the course of a long, strange exchange, a Microsoft chatbot named Sydney professed its incorrigible love for him and tried to persuade him that his marriage was unhappy. Roose promptly published the encounter in The New York Times — Sydney even earned front-page coverage — and the piece became one of the most widely read he has ever written. Unfortunately for Sydney’s admirers, Microsoft swiftly reined the chatbot in, curtailing the very conversations that had produced such exchanges — a lobotomy that drew a barrage of protest, and no small degree of vitriol aimed at Roose himself.

The AGI Chronicles is an absolute delight for the “AGI-minded.” Ilya Sutskever, in one of his famous aphorisms — the so-called “Ilya-isms” — once exhorted his team to “feel the AGI.” Roose’s book, if not quite following in Sutskever’s footsteps, acknowledges that whether one loves it or loathes it, the field is here to stay, and the quest for AGI will continue unabated. The journey is crystal clear; what awaits the passenger at its destination remains a mystery of black-hole proportions. It may be a blindfolded Plutus, cornucopia in hand, blessing the arrival — or it may be a vengeful Hades, delivering the harshest of verdicts upon irreverent souls. Either way, the AGI train has already left the station.
The AGI Chronicles – one of the best reads of 2026 for me!

The AGI Chronicles: The Inside Story of the Race to Cre¬ate an Arti¬fi¬cial Super¬in¬tel¬li-gence by Kevin Roose is published by Farrar, Straus and Giroux, and will be available on sale beginning October 6, 2026.

Thank you, Net Galley for the Advance Review Copy!