The Met Used OpenAI to Resurrect a 1930s Socialite

A wedding dress is just fabric and thread until you know the woman who wore it. That’s the bet The Metropolitan Museum of Art is making with its latest Costume Institute exhibit, ‘Sleeping Beauties: Reawakening Fashion.’ The museum partnered with OpenAI to build a custom chat experience that lets visitors talk to an AI version of Natalie Potter, a New York socialite who died nearly a century ago. Her 1931 wedding dress is on display, and now her personality is too.

The technical recipe is exactly what you’d expect from a serious institutional collaboration. OpenAI didn’t just point a generic model at Wikipedia. The company worked with the museum’s digital team to curate a dataset from scratch — letters, newspaper clippings, historical documents — essentially building a bespoke knowledge base from primary sources. Then they layered custom instructions on top of their most advanced language model, with museum historians and curators steering the process to make sure the AI stayed true to the historical record. It’s a sharp reminder that the most compelling AI applications aren’t about raw model capability. They’re about the data you feed it.

This is a far cry from the chatbot gold rush of 2023, when every brand slapped a generic GPT wrapper on their website and called it innovation. The Met’s approach is deliberately narrow. Natalie Potter isn’t supposed to opine on modern politics or help you debug Python code. She exists within a tightly bounded historical context, with the same safety guardrails as ChatGPT humming in the background to prevent the experience from going off the rails. That constraint is the whole point.

What’s genuinely interesting here isn’t the technology itself — large language models have been able to roleplay historical figures for years. It’s the institutional buy-in. Major museums move slowly and are famously protective of their scholarly credibility. The Met putting its imprimatur on an AI-powered exhibit signals that cultural gatekeepers are beginning to see generative AI less as a threat to authenticity and more as a curatorial tool. Whether museum-goers actually want to chat with a simulated socialite instead of just reading a placard is an open question. But the experiment is worth watching.