ASUC senators and officials have put forward nearly a hundred resolutions flagged for AI-generated or altered writing by the AI detector, Pangram.
The Daily Californian scanned 773 ASUC Senate resolutions from the 2018-19 academic year through the 2025-26 academic year. Of those resolutions, 96 were detected as fully or partially AI-generated or altered.
Pangram identified 47 of those resolutions as 100% AI-generated or altered.
Pangram claims that, on average, for every 10,000 pieces of human-written text, one will be incorrectly flagged as AI-generated — a false positive rate of less than 0.01%. External studies, including one at the University of Chicago found that Pangram achieved near-zero error rates. However, the software can still produce false results.
None of the 359 ASUC resolutions from the 2018-19 academic year through the 2021-22 academic year were flagged for AI-generated or AI-assisted writing. Resolutions identified as fully or partially AI-generated start popping up in early 2023 — shortly after the release of ChatGPT in November 2022.
Potential AI use in ASUC resolutions has increased over subsequent years. Sixty-two out of the 96 resolutions flagged for AI use were from 2025-26 alone.
That means 42% of senate resolutions proposed last year may have been generated or altered with the use of an AI writing tool.
The Daily Cal contacted primary sponsors of 2025-26 ASUC resolutions flagged as potentially containing AI-generated content. The majority of primary sponsors did not respond to requests for comment.
Of the sponsors who did respond, four denied using AI to write resolutions. Two others denied using AI for writing but said it may have been used for editing purposes.
“As a rule of thumb for the ASUC, AI can only be a tool, not a crutch; we always write our own resolutions,” said ASUC President Margaret Solomon in an email.
ASUC officials noted that most resolutions are collectively drafted by staff members, not solely written by the lead sponsors themselves.
“If our office, at any point, utilized AI, it was only used in an editing capacity or to ensure that the general resolution structure remained consistent,” said 2025-26 ASUC senator Nicole Nuñez-Rivera in an email.
Several sponsors also emphasized that the senate resolutions use a standardized legislative template with similar formats and language and claimed Pangram’s AI detection may be less accurate for formulaic writing.
Some staff members with disabilities have been approved for accommodations to use AI as a writing or editing tool, according to ASUC officials.
UC Berkeley alumna Jennifer Tran, who served as 2025-26 ASUC academic affairs vice president, said she and her staff did not use generative AI for writing but that she did run sentences through ChatGPT and Grammarly Pro to edit her spelling, grammar, wording and sentence structure.
“Navigating academic and professional writing with dyslexia can be difficult, and tools like Grammarly Pro and other generative AI chatbots can significantly help students like myself,” Tran said in an email.