AI homework grading at Berkeley: Fields medalist Richard Borcherds stops grading homework for correctness, seeing 'no point' because of AI

AI homework grading at Berkeley turned into a national talking point after the Daily Californian reported on October 23, 2026 that Richard Borcherds, a UC Berkeley mathematics professor and Fields medalist, no longer grades his students' homework for correctness. After seeing how many students used AI to consistently get top scores, he felt there was "no point" in paying readers to check it, since AI use is now unavoidable. He now gives credit for handing homework in. It's a small grading change with a bigger message about AI and work, coming weeks after OpenAI's claim that its agents had solved a version of the Navier–Stokes problem, which our explainer on OpenAI's Navier–Stokes claim unpacks.

This story is based on the Daily Californian's October 23, 2026 report, Borcherds's comments to the California Post, the Berkeley Academic Senate's guidance on generative AI for instructors, Pew Research Center's February 2026 report on teens and AI, and HEPI's Student Generative AI Survey 2026, checked October 27, 2026.

The short version

• Who: Richard Borcherds, a UC Berkeley mathematics professor and Fields medalist.
• What changed: he stopped paying readers (graders) to mark homework for correctness. He "currently grades on completion if a student hands in their homework," the Daily Cal reported.
• His verdict: it's "not a very good system," but he has no alternative for now.
• Berkeley's rules: AI policy is up to each instructor. The Academic Senate asks for a syllabus statement and says no GenAI detector has been validated.
• The wider picture: 54% of US teens told Pew they had used chatbots for schoolwork; 94% of UK undergraduates told HEPI they use generative AI to support assessed work.

What did Richard Borcherds say about grading homework?

According to the Daily Californian's Audrey Baime, Borcherds made the change "after seeing how many of his students used AI to consistently receive top scores." He felt there was "no point" in paying readers to grade for correctness, as AI use is now unavoidable. He now grades on completion and said it is "not a very good system," but that at the moment he has no alternative.

He spoke in a sit-down interview on Tuesday, October 20, which the paper said followed OpenAI's announcement that AI agents had apparently solved the Navier–Stokes existence and smoothness problem. OpenAI made that claim on September 8, 2026; what is checked and what isn't is set out in our explainer.

Borcherds went further than grading. "If AI continues to improve at present rate, then, in a few years, there'll be nothing human mathematicians can do that AI can't do better," he said. "This is the first time anything like this has happened in mathematics, though." He said that five years ago being a mathematician looked like one of the safest careers in the AI boom, and that he's no longer sure whether to encourage students into the math department. "I have no idea what to advise," he said, adding: "I even asked ChatGPT this exact question, and even ChatGPT said it didn't have any idea what was going to happen."

He also expects "inertia": he wouldn't be surprised if the department stayed largely the same for a couple of years.

He gave more detail to the California Post, in a story updated October 25. "Grading for correctness just adds a huge incentive for them to use AI for homework instead of doing it themselves," he told the paper, adding that it "penalizes students for honesty." He said there is "no easy way to tell the difference between human and AI generated work." According to the Post, he bases grades mainly on in-person exams, where students can't use electronic devices, and considers online exams "a waste of time" because some students use AI on them.

Which course is affected?

Neither the Daily Cal nor the California Post names the course, its enrollment or how much homework counts toward the final grade, and Borcherds's public course pages on the Berkeley math department site reviewed for this story are from earlier years. Those older pages show how he has weighted homework before: one Fall 2003 syllabus planned roughly 40% for homework and said "about half the marks for homework will be based on the number of questions that you made a serious attempt on." That is a past course, not this semester's.

What is UC Berkeley's AI policy for coursework?

There's no campus-wide ban or rule; each instructor decides. The Berkeley Division of the Academic Senate, through its GenAI working group, recommends that faculty "include a clear statement on their syllabus about course expectations" for AI use, and reminds them of "their authority over courses and curriculum." It also warns that "there are no validated GenAI detection tools," and suggests alternatives such as in-person proctored exams or oral components.

The Daily Cal reported that the campus has tried AI detection tools, including Turnitin and Rumi, and decided they aren't adequate to standardize across campus. Anne Marie Richard, Berkeley's assistant vice provost for undergraduate education, told the paper that AI policies can vary widely by course level and focus. Sample AI-use statements are collected in the campus AI Hub.

Borcherds's move fits the Senate's logic: if you can't detect AI on take-home work, grade what you can supervise, which is what he described to the California Post: grades based mainly on in-person, device-free exams. Neither paper said whether his exams changed this semester.

How many students use AI for homework?

Most, by the best available surveys, though none covers Berkeley specifically. Pew Research Center surveyed 1,458 US teens ages 13 to 17 from September 25 to October 9, 2025: 54% said they had used chatbots for help with schoolwork, 10% said they do all or most of their schoolwork with chatbot help, and 59% said students at their school cheat with AI at least somewhat often. HEPI and Kortext surveyed 1,054 full-time UK undergraduates for a March 12, 2026 report: 95% use AI in some form, and 94% use generative AI to support assessed work.

Last verified: October 27, 2026

These are self-reported figures from different countries and age groups, so they shouldn't be combined. Pew's margin of error is plus or minus 3.3 percentage points. Neither survey measures how AI use changes homework scores, which is what Borcherds says he saw.

What does it mean for AI and education?

Borcherds is unusual for saying it so bluntly, not for the problem. Graded take-home work assumes the grader can tell the student's work from someone else's, and the Academic Senate itself says the tools to check that don't exist. His fear about careers echoes a broader debate about which work AI exposes, which our explainer on the IMF's 40% figure cautions is not the same as job loss. And mathematics is where AI's gains have been most visible in 2026: besides Navier–Stokes, the KLS conjecture proofs credit AI with a large share of the work.

What's confirmed, reported and uncertain

• Reported (Daily Californian, October 23, 2026): Borcherds's change to completion grading, his reasons and all his quotes; Berkeley's experience with Turnitin and Rumi; Richard's comments.
• Reported (California Post, October 25, 2026): his comments on grading incentives, detection, in-person exams and online exams.
• Confirmed (primary documents): the Academic Senate's syllabus-statement guidance, instructor authority and the "no validated GenAI detection tools" warning; the Pew and HEPI figures.
• Uncertain: which course or courses are affected, how much homework counts, whether exams changed, and whether other Berkeley math faculty are doing the same.