Stanford students found themselves AI-swapped in campus ads
Three uses of AI landed within a day of each other: an institution that rewrote the identities of the people in its own marketing, a startup betting on the materials that come after silicon, and a platform joining a shared pipeline for catching child abuse material at scale.
Stanford's housing and dining division used generative AI to change the race, gender and body shape of students appearing in its advertising, according to the campus paper that broke the story — and the university has not answered for it. Billy Ramirez, a junior, was on his way back to campus when a friend sent him side-by-side images: in one, a photo he remembered a Stanford photographer taking; in the other, he had been removed and replaced by an AI-generated Black woman, and two students to his right had been visibly slimmed. The Stanford Review, a conservative student publication, reported the edits across Residential & Dining Enterprises marketing material and quoted Ramirez: "I don't agree with Stanford making those choices about how we were represented. Seeing my identity changed and being left out of the picture made me feel, in a way, silenced and erased from a representation that was supposed to include me." Stanford has not publicly addressed the account, and the Review says it will keep monitoring R&DE's advertising.
The mechanics matter less than the norm it breaks. Universities have staged diversity in brochures for decades — a fact NPR documented in 2013 — but that took a photographer, a casting call and a reshoot. Retouching a real student's face in a photo he consented to be in is now a few minutes of prompting, and the only detector that fired here was a student's memory of standing in the frame. Every institution that runs a stock image library, a recruiting page or a dining-hall banner now has the same tool and no explicit rule about the people already in the pictures. The obvious gating question — does a photo release cover a synthetic version of the person who signed it? — has no standard answer. If consent for a portrait is going to survive generative editing, universities will have to write that clause themselves, because nothing in existing release language anticipates it. We went through how these edits get made and how they get spotted in AI 101 — What is a deepfake?.
A Singapore startup raised $12 million to sell chipmakers the equipment to grow semiconductors that are not silicon. Nexstrom's tool, called North Star, grows single-crystal 2D materials — transition-metal dichalcogenides — directly on 300 mm wafers using the foundry workflows that already exist, which is the hard part of a technology the industry has pursued for years. Silicon's limits are why it matters: the microscopic channels etched into chips cannot shrink much further before electrons leak and the power bill runs away, exactly the wrong failure at a moment when the AI buildout wants more efficient accelerators. The seed round was led by Xora Innovation with Foothill Ventures and SEEDS participating, bringing the company to $15 million total.
The timeline is the honest part of the pitch. Chief executive Phoebe Tan says Nexstrom's 12-inch system has just been installed and the company has scaled material quality from 2 inches to 6, with 8-inch wafers due by the end of October; co-founder Lance Li, who led post-silicon research at TSMC, has been working on 2D materials since 2012. Nexstrom expects its equipment to be ready for commercial production somewhere between 2030 and 2035 — and TSMC, Intel and Belgium's IMEC are all working the same problem from the other side. That gap is worth holding onto the next time a chip roadmap is described as a 2027 story. The AI compute crunch will be paid for in silicon for a long while yet.
TikTok's US venture joined Lantern, the cross-platform child-safety program, and is funding a $10 million upgrade to the reporting pipeline underneath it. Lantern was launched in 2023 by Meta, Discord and other platforms to share signals about behavior that violates their rules; by the end of 2025, its participants had shared more than 2 million such signals. TikTok's contribution goes to the National Center for Missing and Exploited Children's CyberTipline, a three-year project intended to deploy search and analysis automation that cuts manual processing times in half and pushes actionable reports to law enforcement faster, with AWS, Microsoft and Palantir also contributing resources. Most of TikTok's competitors were already participants, and the venture — spun out of TikTok earlier this year — has been legally obliged to report suspected child sexual abuse material to the CyberTipline regardless.
The reason this is an infrastructure story rather than a compliance one is where automated detection actually finds leverage. A single platform can flag behavior it can see; an abuser moving between apps is invisible to all of them individually, which is why shared signal networks have become the operating layer underneath content moderation. Lantern's director, Marc Leone, framed the addition as strengthening "coordinated enforcement across the wider tech ecosystem," and that is the accurate description — the useful part of this announcement is not TikTok's membership but the plumbing it is paying to modernize.
What to watch: whether any university publishes photo-release language that covers synthetic edits, and when Nexstrom's 8-inch wafers actually run.
If a school can't use your face the way you approved it, whose consent is the release form actually asking for? Tell us in the comments.
Sources: The Stanford Review · The Daily Caller · Resonate · NPR — A campus more colorful than reality · TechCrunch — Singapore's Nexstrom wants to bring 2D semiconductors to chip fabs · SiliconANGLE — Nexstrom bags $12M to replace silicon with 2D wafers · Nexstrom · TechCrunch — TikTok's US arm joins Lantern · NCMEC — CyberTipline modernization
