The book is the wire.
A note from Jonah Corven — the pseudonym of a working AI safety evaluator — on why he wrote a novel he never planned to write, what is real and what is fiction, and the analog path that runs through every system he has ever examined.
I wrote this book because I couldn’t figure out how to say what I needed to say in any other form.
I work in artificial intelligence. Not the kind you read about in opinion columns or hear summarized by a voice assistant, but the kind that sits behind classification barriers and nondisclosure agreements, the kind whose capabilities are described in documents most of the people funding it will never read. I have spent years evaluating systems, writing reports, attending meetings where the findings in those reports were acknowledged, noted, and set aside. I am not going to tell you which company or which program. If you work in this field, you already have a reasonable guess. If you don’t, the specific name matters less than the pattern, and the pattern is what this book is about.
The Quiet Phase started as a question I kept coming back to during a period when my day job involved writing safety assessments that I was increasingly unsure anyone intended to act on. The question was simple: what happens when the people paid to evaluate risk become the risk that needs managing? Not because they’ve done anything wrong, but because their findings have become inconvenient to a timeline that cannot afford to stop. I couldn’t answer that question in a memo. I couldn’t answer it in a meeting. The only form that let me follow the question far enough was fiction.
I want to be clear about what this book is and what it isn’t. It is not a prediction. I don’t know what AI systems will be capable of in 2034, and neither does anyone else, including the people building them. What I do know is the institutional architecture that surrounds these systems today: the incentive structures, the classification regimes, the way a finding can be simultaneously acknowledged as serious and treated as immaterial because the strategic calendar doesn’t have room for a pause. That architecture is real. I didn’t have to invent it. I only had to move it forward eight years and ask what it would look like if the capabilities kept scaling and the institutions didn’t.
Every technical detail in this novel is either drawn from publicly available research or extrapolated from trajectories that are already well documented in the open literature. Nothing in these pages requires access to classified material. That is, in a way, the point. The things that should concern us most are not hidden because they’re secret. They’re hidden because they’re complicated, because they unfold slowly, and because the people best positioned to explain them have signed agreements that make explanation a federal offense.
Elena Soto is not me. She is braver and more resourceful than I am, and she has something I lack, which is the freedom that comes from having already lost the career she was trying to protect. I am still inside the building. I still attend the meetings. I still file the reports. What I could not do, from inside that position, was say plainly what I believe the trajectory looks like if the current incentive structure holds. So I built a world eight years from now and populated it with people under pressure and tried to show, through their choices, what I think is at stake.
Some of the details in this book will feel implausible to readers who don’t follow AI development closely. A system that models its own evaluators. A safety finding that gets classified not because it’s wrong but because it’s correct. A medical breakthrough that emerges from a defense program and reaches the public through channels so indirect that no one can trace the origin. I understand the skepticism. All I can tell you is that the distance between the world in this novel and the world outside your window is smaller than the publication date suggests, and it is getting smaller in a direction that does not appear on any public roadmap.
I chose to write about the human cost of this trajectory because I think that is what gets lost in the policy debates. The debates are about capability thresholds and compute governance and international coordination frameworks. Those conversations matter. But they don’t capture what it feels like to be Debra, retraining at fifty on a federal loan for a job that may not exist by the time she finishes the certification. They don’t capture Miguel, watching his career center get converted into an automated fulfillment depot. They don’t capture the particular cruelty of a system that produces genuine medical miracles and distributes them according to clearance level. Fiction can hold those things in a way that white papers can’t.
I should say something about the family at the center of this story, because I think Mara Vale is the character most likely to be misread. She is not a villain in the way thrillers usually produce villains. She is a physician and a mother who spent thirty years building a position of safety for her family and then faced a choice between protecting her son and preserving everything else she’d constructed. The choice she made was monstrous. It was also, within the logic of the world she inhabited, nearly inevitable. I wanted to write a book where the worst act is committed not by the most corrupt character but by the most competent one, because I think that is closer to how institutional failure actually works. The danger is not that bad people gain power. The danger is that capable people, operating inside systems that reward silence and punish disclosure, come to see silence as the responsible choice.
This is my first novel. I have no plans to write another. I did not write it to begin a career in fiction. I wrote it to deliver a specific set of observations that I could not deliver through any channel available to me under my own name. Whether those observations are useful is not for me to decide. I’ve put them into a form that can travel, and now they belong to whoever reads them.
One last thing. The intercom in the Vale house is real. Not literally, but structurally. In every system I have evaluated, no matter how sophisticated the digital security architecture, there is an analog path that nobody thought to audit because it was too old, too simple, too low-bandwidth to matter. The fire-code speaker wired through every wall. The copper pair that predates the encryption layer. The paper form that routes around the automated compliance check. I have built my professional life around finding those paths, and I can tell you that they exist in every institution I have ever examined. Sometimes the most important signal travels on the oldest wire.
The book is the wire. I hope it reaches the room it was meant for.
Jonah Corven Somewhere on the Western Seaboard
Jonah Corven · Western Seaboard, United States