# A major AI lab has already secretly achieved AGI and is hiding it

**No verdict.** AI progress in the 2020s was genuinely fast and the labs are genuinely secretive, but no verifiable evidence shows any lab has crossed the threshold to general intelligence, let alone hidden that it had. 'No public proof' is not itself proof of a cover-up.

Category: Science, Space & Technology · Era: 2020s · First circulated: 2023 · Believed by: A vocal minority online, spiking around each major model release
URL: https://theconspiratory.com/theory/secret-agi

## Summary
The belief that a leading AI lab has quietly built artificial general intelligence, or even a conscious machine, and is concealing it from the public, regulators, or markets to avoid panic, regulation, or the loss of a competitive edge. It reads the 2023 OpenAI board crisis, the 'Q*' rumors, and executives' own dramatic statements as signs that a hidden threshold has already been crossed. The rapid progress is real; the secret breakthrough is not established.

## The claim
That one of the major artificial-intelligence laboratories has already, privately, achieved artificial general intelligence (a system matching or exceeding human capability across essentially all cognitive tasks) or even machine consciousness, and is deliberately concealing that fact, whether to prevent public panic, to avoid regulation, or to preserve a commercial and strategic advantage over rivals.

## Origin and timeline
- 2022-11-30: OpenAI releases ChatGPT, and a conversational system fluent enough to feel general reaches the public overnight, resetting expectations for what software can do and seeding the intuition that a real threshold is near.
- 2023-03-14: OpenAI releases GPT-4, which posts strong scores on professional and academic exams. A widely circulated Microsoft Research paper calls its abilities 'sparks of artificial general intelligence,' a phrase believers later read far more literally than its authors intended.
- 2023-05-16: OpenAI CEO Sam Altman testifies to a US Senate Judiciary subcommittee, calling for licensing of the most powerful AI systems. Executives publicly warning that their own technology is dangerous becomes a recurring feature of the era.
- 2023-10-30: President Biden signs Executive Order 14110 on 'Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence,' the most sweeping US governance action on AI to date; days later, 28 countries and the EU sign the Bletchley Declaration on frontier-AI safety.
- 2023-11-17: OpenAI's board abruptly fires Altman, saying it had lost confidence in his candor. Five days of chaos follow; he is reinstated on November 22 with a largely new board. The board never publicly specifies what triggered the removal.
- 2023-11-22: Reuters reports that before the ouster, staff researchers had written to the board about a breakthrough codenamed 'Q*' that could reportedly solve grade-school math problems. Online, the vague report is rapidly reinterpreted as evidence the board had glimpsed a hidden AGI.
- 2024-2026: The 'Q*' work surfaces publicly as the 'Strawberry' reasoning model and then the o1 series, an advance but not an AGI. Executives escalate their claims (Altman writes the company is 'confident we know how to build AGI'), and the secret-breakthrough theory spreads with each release.

## The evidence, claim by claim
- Claim: AI capabilities advanced startlingly fast in the 2020s, faster than most experts had forecast.
  Evidence: Confirmed, and not seriously disputed. In roughly two years the public went from novelty chatbots to systems scoring well on bar exams, coding contests, and graduate-level science questions, then to 'reasoning' models that work through problems step by step. Forecasters repeatedly had to pull their AGI timelines forward. This genuine, documented acceleration is the solid foundation the rest of the theory is built on: the mistake is treating fast progress toward a goal as proof the goal has secretly already been reached.
- Claim: The November 2023 firing of Sam Altman proves OpenAI's board had glimpsed a dangerous hidden breakthrough and tried to stop it.
  Evidence: Overstated. It is documented that the board fired Altman citing a loss of confidence in his candor, and that Reuters reported a staff letter about a result codenamed Q*. But the board never named Q* as its reason; later reporting and an internal review attributed the crisis to breakdowns in trust and communication, not to a concealed AGI. Q* itself was described as solving grade-school math, a real advance that publicly became the o1 reasoning models, not a general intelligence. The dramatic timing invites the inference; the specifics do not support it.
- Claim: Executives' own dramatic statements about imminent AGI show they know a threshold has secretly been crossed.
  Evidence: This cuts both ways, which is the problem. Leaders including Altman and Anthropic's Dario Amodei have made striking claims about human-level AI arriving within years. But those are public statements aimed at investors, recruits, and policymakers, and the same competitive and fundraising incentives that could motivate hiding a breakthrough also reward overselling one that has not happened. Notably, the claims are framed in the future tense ('we know how to build AGI,' not 'we have built it'), and more cautious insiders such as DeepMind's Demis Hassabis argue current systems still lack capabilities that a general intelligence would need.
- Claim: A lab has already, definitively achieved AGI or machine consciousness and is concealing it.
  Evidence: Not established, and there is no verifiable evidence for it. No credible leak, document, demonstration, or whistleblower has established that any system has crossed into general intelligence, and today's models, however broad, still fail in ways that narrow systems do (they confabulate, cannot reliably learn continuously, and lack robust real-world autonomy). Corporate secrecy around frontier models is real, but it is fully explained by ordinary trade-secret and safety concerns; secrecy about how a system was built is not evidence that a hidden AGI exists inside it. Absence of public proof is not proof of a cover-up.

## Why people believe it
- The core observation is true: capability really did jump, fast, and each release outran the last, so extrapolating the curve to 'they must be further ahead in secret' feels less like a leap than a natural next step.
- The labs really are secretive. Frontier model details, training data, and internal capabilities are closely held for commercial and safety reasons, and genuine opacity leaves a vacuum that a hidden-breakthrough story fills more satisfyingly than 'they are keeping trade secrets.'
- The 2023 OpenAI board crisis was a real, dramatic, still-only-partly-explained event, and unexplained corporate chaos at the leading lab is fertile ground: if even insiders will not say exactly what happened, a reader can supply the most cinematic possible reason.
- The executives themselves talk this way. When the people building the systems publicly compare their work to a new form of intelligence and lobby for it to be regulated like a weapon, taking them at their most literal and assuming the breakthrough is already here feels like listening, not paranoia.
- Science fiction primed the pattern for decades: the secret, sentient machine hidden by a powerful institution is one of the genre's oldest scripts, so a real technology that looks the part slots into a narrative the audience already knew by heart.

## Open questions
- There is no agreed definition of AGI, or any accepted test that would settle whether a system had reached it, which means the central claim is unusually hard to falsify. Labs, researchers, and executives use the term to mean different things (a system that beats humans at most economically valuable work, a system with human-like general reasoning, a system that can improve itself), and until 'the threshold' is something specific enough to measure, arguments about whether it has secretly been crossed can run indefinitely without resolution.
- The incentives genuinely point in two directions at once, and that ambiguity is not resolvable from outside. A lab that had a dramatic breakthrough would have real reasons to keep it quiet (competitive advantage, avoiding a regulatory clampdown, preventing misuse), and a lab that had not would have real reasons to imply it did (funding, talent, market position). Both pressures are documented and active; the public record does not let an outsider determine which one is shaping any given announcement.
- Exactly what caused the November 2023 removal of Sam Altman, and how large the internal disagreement over capabilities and safety actually was, has never been fully disclosed. Independent reviews and reporting point to trust and governance rather than a hidden AGI, but the board's own detailed reasoning remains substantially private, and that unfilled gap is precisely what the theory expands to occupy.
- How much genuine capability the leading labs hold internally, ahead of what they release publicly, is not knowable from outside and is plausibly nonzero. Companies routinely test more advanced systems for months before release, so some private lead is almost certain; the open question is whether that ordinary gap between internal and shipped models is ever mistaken for, or could ever quietly become, something categorically different from the systems the public can see.

## Sources
- Exclusive: OpenAI researchers warned board of AI breakthrough ahead of CEO ouster, sources say, Reuters, via CNBC (Anna Tong, Jeffrey Dastin & Krystal Hu) (2023): https://www.cnbc.com/2023/11/22/sam-altmans-ouster-at-openai-precipitated-by-letter-to-board-about-ai-breakthrough-sources-tell-reuters.html
- Removal of Sam Altman from OpenAI, Wikipedia (timeline and cited reporting on the November 2023 crisis) (2024): https://en.wikipedia.org/wiki/Removal_of_Sam_Altman_from_OpenAI
- OpenAI releases o1, its first model with reasoning abilities (formerly 'Strawberry' / 'Q*'), Fortune (2024): https://fortune.com/2024/09/12/openai-new-ai-model-strawberry-o1-chatgpt/
- Reflections ('We are now confident we know how to build AGI'), Sam Altman (personal blog) (2025): https://blog.samaltman.com/reflections
- How OpenAI's Sam Altman Is Thinking About AGI and Superintelligence in 2025, TIME (2025): https://time.com/7205596/sam-altman-superintelligence-agi/
- AI luminaries at Davos clash over how close human-level intelligence really is, Fortune (on Hassabis, Amodei & LeCun at the 2026 World Economic Forum) (2026): https://fortune.com/2026/01/23/deepmind-demis-hassabis-anthropic-dario-amodei-yann-lecun-ai-davos/
- Oversight of A.I.: Rules for Artificial Intelligence (hearing record, S.Hrg. 118-37), U.S. Senate Committee on the Judiciary / Congress.gov (2023): https://www.congress.gov/event/118th-congress/senate-event/LC71543/text
- Executive Order 14110: Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence, Federal Register (The White House) (2023): https://www.federalregister.gov/documents/2023/11/01/2023-24283/safe-secure-and-trustworthy-development-and-use-of-artificial-intelligence
- The Bletchley Declaration by Countries Attending the AI Safety Summit, 1-2 November 2023, GOV.UK (2023): https://www.gov.uk/government/publications/ai-safety-summit-2023-the-bletchley-declaration/the-bletchley-declaration-by-countries-attending-the-ai-safety-summit-1-2-november-2023
- Artificial Intelligence Risk Management Framework (AI RMF 1.0), National Institute of Standards and Technology (NIST AI 100-1) (2023): https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-ai-rmf-10

Rated by The Conspiratory, a neutral, sourced encyclopedia of conspiracy theories. Full page: https://theconspiratory.com/theory/secret-agi