“Anthropic co-founder Jack Clark told economists and policymakers at the IMF and World Bank's annual meetings that the singularity, the point where AI speeds up discovery faster than humans can follow, has 'now begun.'”

Clark published the full text of the speech himself, three days after delivering it. In it he writes that researchers and science fiction authors have long forecast AI drastically increasing the pace of scientific and technical progress, and then: "They call this the singularity. I believe that this year we have gained convincing evidence that it has now begun." He titled the newsletter issue carrying the speech "In which I tell the IMF the Singularity has begun," and wrote in his introduction that while people may quibble over details, he thinks it is evident we are in the early stages of the singularity, and that he wanted to use the position of giving the Per Jacobsson lecture to tell the assembled audience of economists, bankers, and policymakers to expect significant, rapid change. Separate accounts of the lecture report that he also said, more plainly, "The singularity is here." The venue is confirmed. The IMF's own schedule lists the Bangkok flagship seminar on AI with Clark participating on 16 October, and the lecture itself was the 2026 Per Jacobsson Foundation Lecture delivered during the IMF and World Bank annual meetings in Bangkok. Reuters covered it: Clark said international cooperation was needed to set "unusual and serious rules" for the technology and that policymakers had a shrinking window to act, and Reuters characterised his remarks as saying growing evidence points to a "singularity" in which AI development outpaces human control. On the benchmark figures, the speech text states: when Humanity's Last Exam launched in January 2025 top scoring systems obtained 8%, and as of that week the highest scoring system was Claude Opus 5.5 with 67.7%, and on FrontierMath, a test released in November 2024 built from hundreds of challenging problems curated and evaluated by professional mathematicians, "Now, they get 94%." Epoch AI's own benchmark index lists FrontierMath Tiers 1-3 (v2) with a highest score of 94%, so that figure matches the benchmark operator's board. On the scientist analogy, the speech says: UNESCO estimates roughly 10 million scientists worldwide, and even with 10 million there are too few human scientists to oversee the work, too few laboratories for materials science, too few biological labs to develop and test drugs, and too few large pieces of scientific equipment, followed by "What happens when instead of ~10 million scientists, we have a billion virtual scientists directed by our own ten million? Or ten billion? That is the world that is coming right for us." On the regulation window and the intelligence ceiling, reporting of the lecture states Clark said humanity has "a handful of years, less than a decade easily," to build serious rules for AI and that Anthropic's chief executive Dario Amodei has said in a public post that it may make sense to limit the ceiling of intelligence allowed in these systems, a notion Clark called a little at the edge of the Overton window. Financial press coverage also rendered the window as a "small number of years". On the $44 trillion figure, Reuters reported one of the company's internal models forecasting that under certain conditions AI could increase US economic output by almost 50% by 2030, and Clark said the estimated potential US GDP of $44 trillion, against a current level of just over $30 trillion, "sounds very outlandish, I know". Benzinga described it as an extreme economic scenario modeled by Anthropic economists, a nearly 50% increase from the current $30.77 trillion in four years, drawn from an Anthropic economic model examining an extreme, "singularity-esque" scenario.