A new frontier model ships every 10 days, and no university is built to keep up

Three labs shipped 28 models in 282 days

The count leaves out image, audio, video, embedding and robotics models, and it leaves out every other lab, DeepSeek included. Google’s changelog lists 39 launches of all kinds. The true figure is higher than 28.

On three days two labs shipped at once: 5 February, 3 March and 28 May. Counting days and not models, a new model arrived on 22 separate days, one every 13 days.

Anthropic’s year shows the pace inside one company.

Opposition on campus is organized and growing

The last two rows give the direction. In the United States, the group turning against AI fastest is the group most common on a faculty. Over the same three years, concern among conservative Republicans fell 14 points. Cal State faculty petitioned their chancellor to end the OpenAI contract, and hundreds of faculty and students at the University of Colorado signed a letter against theirs.

One survey looks like good news and should be read with its definition in hand. EDUCAUSE asked 1,960 people who work in higher education and reported that 94% had used AI tools for work. It counted any software that “has some features that are powered by AI (e.g., Canva, Acrobat),” so a person who opened a PDF is in the 94%. Only 54% knew of a policy covering that use, and respondents ranked “AI’s pace of change” first among their challenges.

Wharton went all-in on AI. Yet one of its data engineers had told developers not to buy the hype.

A school can declare itself all-in months after one of the people who build its research tools went on record as a doubter. A researcher at any university should assume the same split exists at home.

Companies see it too. In a survey of 2,400 knowledge workers, 29% admit to actively undermining their employer’s AI rollout, and among Gen Z workers it is 44%.

Researchers in China, Japan and Korea post their paper pipelines in public, with agent counts, token costs and the checks they keep for themselves. The posts are how-to guides. The Western letters are refusals.

AACSB’s 2026 standards become mandatory in 2027-28 and do not contain the word AI. Nothing above the university is pushing it to move faster.

Make noise, in writing, this week

1. Email the CIO and your dean. Ask which current model you may use, on which data, and by what date you will have an answer.
2. Name the model and its date. “Opus 5.5 shipped on 22 September 2026. When can I use it?” is harder to file away than “what is our AI policy?”
3. Ask the library for the AI clause in each data license. Ask in writing, and ask what it would take to change it at renewal.
4. Ask who owns AI policy for research. If nobody does, say so at the next faculty meeting.
5. Copy your colleagues. One request is a ticket. Ten are an agenda item.
6. Do not wait for the answer to start. Pay for your own plan and build on public data until the answer comes. The SEC filings guide needs no license.

Every request above is inside the rules. A data license still binds you while you wait, and the licensed data guide shows how to keep working without breaching one.