“Credit the AI tool”:
Stanford’s AI Guidelines Page

Stanford’s statement says it confirmed that AI was used to alter an image on its banners, and that those banners have been removed. As of October 20, 2026, none of the files this edition read names the AI tool used on the banner, prints any prompt, or names an individual or position that made, ordered or reviewed the edit.

By William Douglas Baker

Founding Editor

Stanford said in a statement, printed in full by ABC7 San Francisco in a report dated September 24, 2026, that it had “reviewed the matter and confirmed that AI was used to alter the image in violation of university policy.” The statement adds: “The banners with the image in question have been removed.” The Stanford Daily reports that Stanford confirmed Residential and Dining Enterprises (R&DE) used generative AI to modify the appearance of several students in a promotional image. The Daily says it confirmed that the photo was used in a custom banner in the Governor’s Corner student housing center, which has since been taken down. The Daily states that R&DE changed the students’ clothing into Stanford merchandise, slimmed down two students’ faces and “entirely changed one student’s race, gender and appearance.” In every printing this edition read through October 20, 2026, Stanford’s statement does not say why the edit was made, and this edition does not guess.

The SF Standard states that Billy Ramirez, a senior, was the student who was replaced in the photo. Ramirez told the Daily that a friend initially sent him the edited photo. “I thought that maybe they merged me with the guy in the middle, but I was like ‘Oh, no, they actually replaced me,’” he said. The rest is about documents.

The page the Los Angeles Times linked

In the Los Angeles Times, which quotes an email from Stanford spokesperson Charlene Gage, the words “AI policy” in her sentence “Stanford’s AI policy is clear:” link to the Stanford University Communications page “AI guidelines for marketing and communications.” As captured by the Internet Archive on June 9, 2026, it says: “Prompts and outputs are discoverable. Treat prompts and AI outputs as records that may be discoverable or subpoenaed, just like your emails.” The Internet Archive capture of September 23, 2026 and a direct fetch of October 14, 2026 carry the same sentences word for word.

On images, the page says: “AI-generated images or media: Credit the AI tool in the appropriate credit line, as you would credit any other production tool.” For published content, the page gives an example disclosure: “This article was supported by research using [tool name].” The page closes with a disclosure: “Consistent with these guidelines, this document was supported by research using Anthropic Claude. All outputs were reviewed by humans and the final expression was produced by the authors.” The page names the AI tool it says supported research for the page itself. Through October 20, 2026, no file this edition read names the tool used on the banner or dates the alteration, so the files do not establish whether the June 9, 2026 text was on the page when the image was altered.

Where the statement and the page meet

The statement, as ABC7 prints it, also says: “Stanford’s AI policy is clear: The use of AI in producing or altering images of Stanford people, events, research, facilities, or achievements is strictly prohibited. Both the alteration and lack of disclosure in this case violate that policy.” In the four Internet Archive captures (January 7, June 9, September 23 and September 24, 2026) and the one direct fetch (October 14, 2026) this edition searched, the guidelines page does not contain the phrase “strictly prohibited.” Among activities requiring the highest level of care, the page’s care column reads “Not permitted without explicit consent” for “Any application involving the creation of synthetic media depicting real individuals.” Among activities requiring a higher level of care, the page lists “Using AI to generate or substantially enhance images or video for public use,” with the care column reading “Ensure compliance with intellectual property guidelines and credit the tool.”

The two texts sit side by side, and no conclusion is drawn from the difference in wording. The page also says: “Finally, the absence of a specific use case in this document should not be read as a prohibition.” This edition applies none of the page’s sentences to the banner or to R&DE. Through October 20, 2026, the files this edition read do not establish whether any of the three students pictured gave permission, written or otherwise, for the photograph’s use in the banner or for the AI alterations.

A demand, and what the files do not say

A letter dated September 24, 2026 on the anonymous Substack The Other Two, whose authors say they are the two other students pictured, demanded “a transparent and thorough investigation regarding the image’s creation, including the prompt and the bureaucratic processes it had to go through.” The word “prompt” is theirs. This edition does not say that one exists. As of October 20, 2026, none of the twelve article files this edition read, or the five page files it read or searched, names the AI tool used on the banner or prints or describes a prompt.

None of those files says whether a prompt or log was kept, names an individual or position that made, ordered or reviewed the edit, or reports any further findings of the review Stanford’s statement says it conducted. The Daily names the unit, R&DE, not an individual or position. Not read: The New York Times, the San Francisco Chronicle, The Washington Post, NBC, People, Fox, CNN, Inside Higher Ed, Business Insider, Stanford’s newsroom and R&DE pages, Fizz posts, and the Daily’s September 24, 2026 article on the other two students.

What NPR reported in 2013

The Los Angeles Times writes that “generative AI tools are making it faster and easier for anyone to create false representations.” NPR reported, in an Associated Press image caption, that “University of Wisconsin officials added the face of a black student, Diallo Shabazz, to a file photo for the cover of the school’s 2000 application booklet.”

Shabazz told NPR that an admissions counselor walked up to him and said, “Diallo, did you see yourself in the admissions booklet? Actually, you’re on the cover this year.” “So I flipped back, and that’s when I saw my head cut off and kind of pasted onto the front cover of the admissions booklet,” he says. On our reading, each man’s own account has him learning of the altered image from another person: a friend for Ramirez, an admissions counselor for Shabazz. Stanford’s statement does not say why the edit was made.

Our reading, in this comparison only: in the 2000 Wisconsin case and the 2026 Stanford case, an institution altered a photograph of students for its own promotional material, and the student at the center of each edit learned of it from another person. The methods differ: a pasted head, in Shabazz’s account, and AI, in Stanford’s. What the Stanford case adds is a document: the page says to treat prompts and AI outputs as records, and the anonymous authors who say they are the other two people pictured demanded an investigation of the image’s creation “including the prompt.”

If an organization features you

Our analogy, not a finding: any organization that features you in its own materials publishes its own copy of your likeness, and, by Stanford’s own statement, such a copy can be altered with AI and put on banners.

Where the record stops

This edition states no cause and no motive, and does not say that any organization that features you has altered your likeness. Absence from the files this edition read is not evidence that a prompt, log or reviewer does not exist. The gap is measured, not filled.

Stanford’s statement was read as outlets printed it, not on a Stanford page, and Stanford’s own records were not read. Disclosure: this story ran below this publication’s usual bar for selecting a story, because an institution altering a student’s image in its own promotional material has a precedent and the person replaced is a student, not one of the professionals this publication writes for.

Declassified is produced with AI tools that include Claude, made by Anthropic. The guidelines page was last fetched directly on October 14, 2026; NPR’s 2013 report was read as text, its audio not heard. We map the record, not any person or company, to Digital Derangement Syndrome™ (DDS™), the house name for the condition in which AI systems fail to recognize real expertise. That mapping is editorial, not any source’s finding.