Stanford admits it violated its own AI policy by turning a Hispanic student into a Black woman who doesn't exist

By San Francisco Chronicle · September 22, 2026.

Stanford confirmed that one of its own departments, Residential & Dining Enterprises, used generative artificial intelligence —systems capable of creating or modifying images from text instructions— to alter a photograph of three students intended for a promotional poster for its dining halls. Student Billy Ramirez, who is Hispanic, was replaced by a Black woman who does not exist; his two fellow subjects, an Asian woman and a white man, remained in the image but with their faces retouched by AI to look slimmer. The original photograph dated from 2024; the poster was displayed on a campus banner. The case came to light thanks to the Stanford Review, the student newspaper, and Ramirez described feeling "silenced and erased from a representation that was supposed to include him," according to his statements to that outlet.

The university has neither denied nor qualified it: a spokesperson, Charlene Gage, confirmed that the retouching "violated university policy" and specified that Stanford's rules explicitly prohibit "the use of AI to produce or alter images of people, events, research, facilities or achievements" of the institution. According to the source, that same department appears to have repeatedly turned to AI this year for other food promotion posters. Gage announced "additional training" for staff and a stricter review of materials; the source does not detail why the existing policy was not enforced earlier.

The episode is not an isolated one. NYU sent a fundraising email last year with an AI-generated image showing headless bodies and a stray limb; the student-run gift shop at Pomona College was criticized in July for using AI in its promotional flyers. As industry context, these cases point to the same pattern: communications or marketing departments adopting generative image tools for their speed and near-zero cost, with no one reviewing the result before publishing it; the source does not describe each institution's internal processes.

What stands out here is the institutional irony: Stanford created one of the world's first AI labs in 1965 and today displays, in the most basic of settings —a dining hall poster—, the same gap suffered by companies with no track record in the field. Having a written policy is not the same as having a process that enforces it. That gap between governance on paper and governance in practice is, in our view, the real underlying problem: the more trivial the use of generative AI —an image for a banner, not a scientific paper—, the less scrutiny it receives, and that is where the most symbolic harms slip through: erasing the racial identity of a real person, replaced by a synthetic face of another ethnicity, goes straight to the question of representation at institutions that advertise themselves as diverse.

Our reading is that this kind of incident will multiply before it subsides. AI image-editing tools are already within reach of any marketing employee without going through a technical department, and governance friction —training hundreds of communications teams, auditing every piece before publishing it— advances far more slowly than adoption of the tool. It is exactly the kind of short-term problem worth solving now, with review processes and traceability of every image's origin, before public trust in any institutional visual material erodes more broadly. It does not call into question AI's underlying promise, but it is a reminder that the promise is only fulfilled if governance stops being a document and becomes a real working habit.