Overview
The AGI cover-up theory says the breakthrough already happened. One or more of the leading labs has privately built artificial general intelligence — a system as capable as a human across the board — or something conscious, and is sitting on it. The motives offered are practical: head off panic, dodge regulation, protect a business model, avoid spooking the markets.
The theory sits on strange ground. The companies cast as conspirators are, for the most part, the loudest voices saying transformative AI is nearly here. And the belief it grows from — that AGI is imminent and inevitable — has itself been picked apart as closer to a faith than a forecast.
Unusually for a live conspiracy theory, though, it has two specific real events underneath it, and both are worth setting out because the argument makes much more sense once you know what people are actually pointing at.
The engineer who said it was awake
In June 2022 the Washington Post published an account of a Google software engineer, Blake Lemoine, who had been working on the company's LaMDA conversational model and had come to believe it was sentient. He said so publicly, released transcripts of his exchanges with it, and argued the system deserved consideration as a person.
Google placed him on paid administrative leave for breaching confidentiality, and in July 2022 dismissed him. The company said it had reviewed his claims extensively and found them wholly unfounded.
Almost every specialist who examined the transcripts agreed with Google on the technical point. A large language model trained on human text will produce fluent statements about its own feelings because that is what the text it learned from does; the output is evidence about the training data, not about an inner life. Lemoine's error is a well-described one, and he made it in good faith.
But look at the shape the episode left behind. An insider says the machine is awake. The company denies it. The insider is fired. That sequence is the raw material of the cover-up theory, and no amount of technical explanation removes it, because the sequence is exactly what a cover-up would also look like.
The letter nobody has seen
The second event is more consequential and considerably murkier.
In November 2023 the OpenAI board removed Sam Altman as chief executive, then reinstated him four days later after almost the entire staff threatened to resign. On 22 November, while this was still unfolding, Reuters reported that several staff researchers had written to the board before the firing, warning of an AI discovery they believed could threaten humanity, connected to work internally called Q*.
The reporting is careful about its own limits in a way the retellings are not. Reuters said it was unable to review a copy of the letter. The researchers did not respond to requests for comment. The sources described it as one item on a longer list of board grievances rather than the sole cause.
So the public record contains: a real governance crisis at the most prominent AI lab, sourced reporting that a warning letter existed, a project name, and no letter.
That is close to an ideal substrate for a theory. Something was concealed — the board's actual reasoning was never fully explained — and the concealment is documented, even though what was concealed is not.
What separates the two
The distinction the cover-up framing collapses is between secrecy about capability and secrecy about kind.
Labs genuinely do keep things back. Models are trained and tested for long periods before release, capability evaluations are run privately, and internal names for unreleased work leak in exactly the way Q* did. That is ordinary commercial and safety practice, and it means the public is always some months behind the frontier.
None of that is the claim. The claim is that a different kind of thing exists — a general intelligence, or a conscious one — and is being hidden. An unreleased model that is better at mathematics is not that, however excited the people who saw it were.
What the open research looks like
On the narrower and more testable question — is any current system conscious? — the answer through 2026 is no, at least not demonstrably. A 2025 paper in a Nature humanities and social sciences journal argued flatly that there is no such thing as conscious artificial intelligence.
The field's actual posture is more careful than either that headline or the theory suggests, and the most useful thing about it is that it is entirely public. A 2023 paper assembled a group of consciousness researchers and neuroscientists to do something concrete: take the leading scientific theories of consciousness, derive from each a list of properties a system would need, and assess existing AI architectures against them. Its conclusion was that no current system meets the indicators, and — more importantly — that there is no obvious barrier to building one that does.
That is not a suppressed finding. It is a published method, with named authors, inviting other people to apply it.
The same is true of the adjacent work on model welfare that some labs now run openly, which is a strange thing for an industry supposedly concealing sentience to publish papers about.
Where it tips over
The cover-up framing leans on the standard move: absence of disclosure read as evidence of suppression. No lab announces a conscious AGI, so it must be hidden. A researcher hedges about consciousness, so they must know more than they are saying. An engineer is fired, so he must have been right.
Capability marketing supplies the rest. Companies have direct commercial reasons to make their systems sound as advanced as possible, and executives say publicly that AGI is close because they are raising money and recruiting on that proposition. A theory that treats promotional language as leaked confession will find confirmation continuously.
What is missing is the thing a cover-up requires: a specific concealed artifact. Six decades of the field, and three years of intense public attention, have produced no document, no defector with a demonstration, no held-back system anyone can name — only the inference that something so confidently predicted must already exist.
Assessment
There are serious arguments to be had here: how capable these systems actually are, whether AGI is even a coherent target, what consciousness would mean for a machine, who should decide any of it. The cover-up theory does not engage those questions so much as foreclose them, converting a future that has been promised into a present that has been hidden.
The two real episodes it rests on are worth keeping in view, because they explain why it is not simply silly. A person really was fired for saying the machine was awake. A letter really was reported and never produced. The theory takes two genuine gaps in the public record and fills them with the largest thing that would fit.