AI100 Report Says AI Has Gone Mainstream in Five Years

Estimated reading time: 8 minutes · Last updated: 10 November 2026

Generative AI has moved from a niche research tool to something millions of people use every day. That is the central finding of the 2026 AI100 report, the third assessment from the One Hundred Year Study on Artificial Intelligence. The panel, administered by Stanford University's Institute for Human-Centered Artificial Intelligence (HAI), traces the shift to five years of training models on ever larger amounts of compute and data. It now sees that approach running into limits. Titled AI Goes Mainstream: The Scaling Era and the Rise of Generative AI, the AI100 report sets out 12 headline findings. They cover science, investment, warfare, synthetic media, public participation and governance. It also checks today's picture against the panel's 2021 edition.

What we did not foresee was the 'ChatGPT moment' that took the world by storm.

Iyad Rahwan, member of the 2026 AI100 Study Panel

Why the AI100 Panel Sees Scaling Running Out of Road

For five years, the panel argues, the main engine behind generative AI has been scaling: training systems on steadily more computing power and more data. That recipe produced the leap that put chatbots in front of the general public. The panel now says it is meeting hard limits on both inputs.

Its answer is a shift toward reasoning and inference-time compute. Inference-time compute is the processing a model spends while it works out an answer, as opposed to the processing spent training it. The panel presents this as the direction the frontier is moving, not as a finished transition.

Iyad Rahwan sat on the 2026 panel and directs the Center for Human and Machines, part of the Max Planck Institute for Human Development. In his view, the 2021 edition had anticipated rapid advances yet failed to see one development coming. "What we did not foresee was the 'ChatGPT moment' that took the world by storm," he said. Billions of people now turn to AI on a regular basis, he added, and that has shifted how business and industry regard it.

The panel's last finding is about forecasting itself. Almost nobody foresaw the present state of AI five years ago, it says, and if development keeps its current speed, the period to about 2031 should prove just as hard to call.

AI Has Reached Fields That Thought They Were Safe

The panel compares AI with the World Wide Web as a disruptive technology. It names the creative industries, mathematics and software development as fields already being reshaped, each of which had assumed it lay out of reach. They join journalism, music and retail on a list of sectors that technology has upended over the past 30 years.

The panel goes further. It warns that other areas convinced of their own immunity are likely to be disrupted soon. That is a forecast rather than a measurement, and the panel does not say which sectors it expects to be next.

Science gets a finding of its own. The panel says AI is changing research at speed, and points to the 2024 Nobel Prizes in both Physics and Chemistry as formal recognition of that shift. In its account, work that once took years can now be finished in months or even weeks.

Faster research is not presented as a clean win. The panel says AI-driven discovery also opens a set of new scientific grand challenges, and that those problems need attention in their own right.

Who Pays for AI Infrastructure, and Who Controls It

The panel describes investment in AI infrastructure as without precedent in both size and pace. In its assessment, that spending is holding up parts of the U.S. economy and dominating capital markets. It is also feeding worries about a financial bubble and putting pressure on energy and freshwater supplies.

The energy side is already visible. The Max Planck Society, which released Rahwan's comments, points to hyperscale data centres in the Netherlands. Their power needs are pushing up investment in grid infrastructure and renewable energy.

The second concern is control. Frontier AI refers to the most capable systems, those at the field's leading edge. According to the panel, only a handful of actors, based in the U.S. and China, are building them. Internationally, that prompts questions about who will ultimately own these technologies and who will get to use them.

None of this amounts to a verdict that a bubble exists. The panel records the fear, not a crash, and that distinction matters. Its argument is that the money and the power are concentrated enough to become a policy problem whatever markets do next.

Security, Evidence and a Public Left Out of the Decisions

Cybersecurity is singled out as a leading area of risk. The panel sees it both as a field where AI is being abused and as a possible route by which powerful systems could cause harm in the physical world. Military AI, it says, has moved from speculation to a force that is reshaping active conflicts, though no specific wars are named.

Synthetic media gets one of the sharpest findings. Images, audio and video generated by AI have, in the panel's view, ended a century-old assumption: that photographs and recordings can be trusted as evidence by default.

The panel also argues the public has had little influence over AI's development and deployment, even as the technology reshapes the lives of billions. It says risks once dismissed as science fiction are now emerging, while older ethical concerns need more attention, not less.

Governance is described as fragmented. The panel finds it increasingly treated as a national-security matter, with little agreement among states on where it should go.

How a Study Built to Run for a Century Works

Launched in 2014, the One Hundred Year Study on Artificial Intelligence is run by an international Standing Committee of AI experts, with Stanford HAI handling its administration. At five-year intervals the committee assembles a Study Panel, which assesses how far AI has come and what it means for individuals, for communities and for society at large, weighing technical and ethical questions alongside economic and societal ones. Reports have so far appeared in 2016, 2021 and 2026.

The 2026 panel drew on computer science, philosophy, sociology and related disciplines. It answers 12 standing questions about AI's advances, opportunities, dangers and social effects. It adds deep dives into AI geopolitics and the scaling era, and sets recent developments against the 2021 panel's assessments.

That built-in comparison is the study's real strength. A one-off expert forecast cannot be checked against its own record. A recurring panel can, and Rahwan's admission that his predecessors missed the ChatGPT moment shows the method working. The panel's own conclusion is that the coming years will decide how AI's opportunities and risks play out.

What could push AI's trajectory either way

The bottom line

The 2026 AI100 panel describes a technology that has outgrown the expectations of its own previous assessment. Scaling delivered generative AI to millions of users and sped up science enough to feature in two 2024 Nobel Prizes. Yet the same five years produced concentrated control, heavy strain on energy and water, military deployment and media people can no longer take at face value. The panel's sharpest point may be its plainest. Billions are affected, while the public and most states have had little say in the direction. The value of a study designed to run for a century is that its next panel, due around 2031, will be able to judge whether these warnings held.

What to watch

• The next AI100 Study Panel report, which would fall around 2031 if the study keeps its five-year cycle; no exact date has been set.
• Whether new frontier models confirm the panel's view that progress is shifting from raw scaling to reasoning and inference-time compute; no date has been set.
• Any sign of the international agreement on AI governance that the panel says is currently missing among states; no date has been set.