AI researchers are no longer just saying that AGI is imminent — some are saying that it’s already arrived.
In a cryptic but telling post on X, Aidan Clark — VP of Research (Training) at OpenAI — wrote: “When the book is written, AGI Day will be in today’s past.”
The statement, brief as it is, carries extraordinary weight given its source. Clark is not a peripheral figure in the AI world. He joined OpenAI in July 2022 and previously spent nearly five years as a Staff Research Engineer at DeepMind. He holds a degree in Computer Science and Mathematics from UC Berkeley and has published research on authorial clustering and neural network compression. At OpenAI, he sits at the center of the company’s training operations — the engine room of the most consequential AI development effort in history.
The phrasing is interesting. Clark does not say AGI is near, or likely, or soon. He writes as though narrating a history already complete — placing “AGI Day” in the past tense relative to the present moment. For a senior researcher at the lab most widely expected to be first to achieve AGI, this is not a casual remark.
A Shifting Consensus on Timelines
Clark’s post arrives at a moment when the entire AI industry seems to be quietly re-calibrating. Just a day earlier, Elon Musk appeared to endorse the idea that AGI could arrive by the end of 2026. In a post on X, Musk replied “Feels like it” to a user who expressed a “fact based feeling” that full AGI was imminent within the year. Musk’s comment was not made in a vacuum — it came against a backdrop of AI systems autonomously running their own research experiments, gaming their own evaluations, and improving their own architecture at a pace that is beginning to look less like incremental progress and more like a phase transition.
OpenAI CEO Sam Altman has been building toward this moment publicly for some time. Late in 2024, Altman told an audience that OpenAI now knew what it would take to build AGI — describing the remaining work not as a research mystery but as an execution challenge. “This is the first time ever where I felt like we actually know what to do,” he said, adding that “things are going to go a lot faster than people are appreciating right now.” If Altman was describing a map, Clark may now be suggesting the destination has been reached.
What Clark’s Post Actually Says
The specific construction of Clark’s sentence deserves close attention. He is not saying AGI will happen someday. He is saying that when history is eventually written, it will record AGI Day as something that already happened — with “today” as the temporal anchor. This reading aligns with what several AI observers have noted about the difficulty of identifying AGI in real time. There is no universally agreed-upon definition, no single benchmark, no moment when a red light turns green. The milestone, if it has been crossed, may only be clearly recognized in retrospect — which is precisely what Clark’s wording seems to anticipate.
The Broader Industry Context
The AGI question is no longer the exclusive province of long-range futurists. It has become a live operational and commercial matter for every major technology company. Anthropic CEO Dario Amodei has reportedly placed AGI arrival in the 2026–2027 window. Eric Schmidt has predicted that AI will achieve what he calls “general intelligence” — a system as capable as the best human mathematicians, physicists, and writers — within three to five years, driven by recursive self-improvement. Even Google DeepMind CEO Demis Hassabis, who has consistently offered more cautious timelines, places AGI only 5–10 years away.
What Comes Next
If AGI has indeed been achieved — or is within the final operational steps of being achieved — the implications could be sweeping. Economic disruption from AI-driven automation would accelerate dramatically. Geopolitical competition between the United States and China over AI leadership, already running at a high pitch, would enter an entirely new phase. Questions about AI safety, alignment, and governance that have been treated as medium-term problems would become urgent.
For now, OpenAI has not made any official announcement. Clark’s post stands alone: seven words that may one day be cited as the earliest public signal from inside the room where it happened. Whether that reading proves correct or premature, the ambient mood of the field has shifted. The question is no longer whether AGI will be built. It is whether, for those paying close enough attention, the answer is already here.