Stanford pulls a banner after using AI to replace a Hispanic student with a Black woman

By The Washington Post · September 24, 2026.

Billy Ramirez, a senior at Stanford, was on his way back to campus on Friday when a friend sent him two photos: the one taken of him in his freshman year and the version the university had posted that fall on a campus banner. In the second, his face was no longer his own: it had been replaced, through AI editing, with the face of a Black woman. As The Washington Post reports, Stanford took the poster down and acknowledged that the way AI had been used breached its own policy.

The available material stops there. We do not know what tool was used, who commissioned the edit, whether the aim was to visually "diversify" the campus's promotional material, or whether this is an isolated case or the first to come to light. Faced with that scarcity of data, the honest course is not to fill the gaps with conjecture: what is verified is that a top-tier university used AI to alter the racial identity of a real student in a piece of institutional communication, and that once it became public it had to backtrack.

Our read is that the case matters less as an exception and more for what it foreshadows. Generative editing tools have made retouching a photograph so cheap that institutions with serious marketing departments — not amateurs with an app — can end up publishing something that erases a specific person without anyone raising an alarm before the banner goes to print. It is a problem of internal governance, not of the technology itself: what is missing is a quality control as basic as asking the person portrayed whether they recognize their own face.

Broadly speaking, incidents of this kind are the friction cost of the current phase of generative AI adoption in institutional communication: cheap, powerful and still poorly supervised. It does not change our underlying thesis about AI's long-term potential, but it does confirm that the hardest part to get right is not the model's capability, but the human judgment — and the verification processes — surrounding its use.