An AI Detector Flagged a Berkeley Professor's Op-Ed as 33% AI
She had already admitted it — and that is the part no one has a rule for
The flag was real and the detector is a good one. But it caught no secret: the professor had already said she used AI to help edit the piece. What the number cannot settle is the only question in dispute — whether editing an opinion column with AI breaks any rule at all.
How it started
A calculus teacher's warning turned into a viral column
On 15 August 2026 the San Francisco Standard ran an op-ed by Zvezdelina Stankova, a teaching professor of mathematics at UC Berkeley and the founder of the Berkeley Math Circle. She wrote that some of her students arrive "five to eight years" behind, that office hours meant for integrals turn into lessons on fractions, and that the decline accelerated once the University of California stopped looking at standardized test scores.
It was a warning about students who, in her telling, cannot do their own foundational work. It travelled fast, picked up by outlets across the political spectrum. Nothing in the published piece said anything about how it was written.
What changed since
A student paper ran it through a detector, and she answered
Over the weekend, journalists at Berkeley's student newspaper, the Daily Californian, thought the prose read like a machine's. On 18 August the reporter Francis Luo published the result of running it through the AI detector Pangram, which flagged the article as "33% AI-generated or AI-assisted."
Stankova did not deny it. She said she had used AI "to help edit the piece," that the column was "the result of several hundred person-hours of intensive human work … of which about 80 hours are my own," and that a faculty team and several journalists had worked over her drafts for three weeks. She told the Guardian that AI had also helped her locate documents, and that "all analysis is the result of the team members." The admission, not the detector, is where the facts stop being contested.
Where it stands now
The detector is accurate, and it still cannot settle the argument
The flat version, the one we would stand behind, is this: it is neither a hoax exposed nor a detector caught inventing things. Pangram is one of the more accurate tools of its kind — independent benchmarking puts its false-positive rate near zero on passages this long — so the flag was almost certainly not an error. And Pangram's own output separates "lightly" and "moderately AI-assisted" from "fully AI-generated." A 33% assistance score is a reading of editing help, which is exactly what Stankova described, not a claim that a machine wrote a third of her column.
That leaves the question everyone actually reached for a detector to answer: was any of this against the rules? The Standard's stated policy is that "while AI may assist, our expectation is that humans are behind every article we publish and take responsibility for every word." On its face that permits what she described. No published rule at Berkeley, at the Standard, or anywhere in general practice says an opinion writer may not use AI to edit — even as other contexts have started to draw that disclosure line in law. The awkwardness is real; the violation is not established.
What happens next
Nothing is scheduled, and the outlet has added no note
There is no dated next step on the AI question. As of the reporting, the Standard had not appended a disclosure note to the op-ed, and no institution had said it would rule on whether an AI-assisted column must declare it. The subject Stankova was actually writing about has a longer clock — UC has not counted SAT or ACT scores since a 2020 court ruling, and its admissions board has begun re-examining that policy without adopting any change — but on the narrow matter of the AI flag, nothing is on a calendar.
Sources and verification
- The Guardian: Stankova's admission that she used AI to "help edit the piece" and to locate documents; her "orthogonal … distraction" remark; and the San Francisco Standard's statement that "while AI may assist … humans are behind every article … and take responsibility for every word."
- San Francisco Standard: the op-ed itself (15 August 2026), its "five to eight years" and severe-deficit claims, and the absence of any AI note in the published piece.
- The Daily Californian: Francis Luo's 18 August report that Pangram flagged the article as "33% AI-generated or AI-assisted," and Stankova's "about 80 hours are my own" response.
- Cryptobriefing: Pangram's four output bands (Human-Written / Lightly / Moderately AI-Assisted / Fully AI-Generated) and assistance score, its 2023 founding, and its 99.98% detection / 1-in-10,000 false-positive claims.
- GradPilot: comparative false-positive rates (Pangram near zero on medium and long passages, per University of Chicago Booth), and the class-level finding that 2023-era detectors falsely flagged 61.3% of non-native English essays.
- University of California press room: the Board of Regents' 21 May 2020 decision phasing out the SAT/ACT requirement.
- University of California / news.uci.edu: the 2 September 2020 court ruling that UC could no longer consider SAT/ACT scores in admissions.
- Stankova's UC Berkeley faculty page: confirms she is a teaching professor of mathematics at UC Berkeley and founder of the Berkeley Math Circle.