'Everyone has an AI secret': This artist has spent a year collecting and sharing hundreds of anonymous AI confessions
CNBC Β· C Β· trust 54/100

In September 2025, Olivia Tai trekked to Mission Dolores Park in San Francisco with a makeshift box, blank notecards and a cardboard sign with a message scrawled in Sharpie: "Confessions Box: Tell Me Your AI Secret."
She set up her DIY confessions booth on a sidewalk and hid out of sight, waiting for passersby to stop and contribute to her participatory art project.
Tai thought she might get a few notes referencing common use cases, like "I use AI to write my emails." Hours later, after sifting through her small collection of crowdsourced messages, one surprised her:
"I cloned my ex'es [sic] voice to have a last conversation with him. And it helped me finally move on. 10/10 recommend."
In three sentences, Tai felt she knew so much about the writer: What they were going through, how they were processing their grief over a breakup, and how they were able to use technology to bring closure they were seeking but couldn't, or wouldn't, get from a human.
"From that point on, I just felt like I had to keep collecting these notes and seeing what people would say," Tai tells CNBC Make It. Her project, " What We Tell AI ," began to take shape.
Tai, 26, is an artist in San Francisco who facilitates workshops exploring the relationships between humans and artificial intelligence.
She says the most popular part of her workshops involve participants writing down an AI-related confession, dropping the note into a bowl, and taking turns reading the anonymous secrets out loud. It was an instant hit, so Tai took that enthusiasm and, inspired by PostSecret , the 2000s-era community mail art project created by Frank Warren, set out to create a new kind of confessional experience in the age of AI.
About half of U.S. adults now report using AI chatbots like OpenAI's ChatGPT, Google's Gemini or Microsoft's Copilot, up from the 1 in 3 adults who reported doing so in mid-2024, according to a recent survey from Pew Research Center. Today, roughly 1 in 4 American adults use these tools every day.
Dating and breakup confessions are some of the most common Tai receives, she says, as people turn to AI for help on everything from writing their Hinge messages to getting over a situationship or even practicing how to ask someone out.
Tai says she's gotten more secrets than expected about people with AI boyfriends and girlfriends .
Some of the project's AI confessions reveal how people use the tech in their work or schooling, but most revolve around how they use it to navigate interactions with other humans. People are increasingly turning to AI chatbots for questions around health and relationships, according to research from the AI Observatory, a public platform that analyzed over 24,000 annotated AI conversations from seven different platforms from 2023 to 2026.
Tai has taken "What We Tell AI" to more than 20 sites across the Bay Area, New York City, Denver and elsewhere; her confession booth pops up in public spaces like parks and shopping malls, as well as gatherings where AI is the subject, like at conferences and workshops. The project has collected over 350 handwritten notes, according to its website .
Tai says she considers the collection as an "ethnographic snapshot" of a certain place at this point in time, rather than representative of how people feel about AI overall.
While some confessions express a sense of pride in how they use AI to help them through a rough patch, like the writer who needed closure with an ex, others express a sense of shame and guilt about turning to AI for companionship, validation and other help.
"Forgive my sin: I talk to AI way more than to ALL people in my life combined," reads one note from Palo Alto, California, that ends with a drawing of a frowning face.
"I used AI for basically all of my school and feel like I don't deserve my degree," reads one New York submission.
The project has also yielded a few messages from people who don't use the tech: "I've never used AI. I'm scared to contribute to our own downfall," reads a confession from San Francisco.
Some have challenged Tai's own assumptions: In San Francisco, where she expects more pro-AI tech workers, she once got a note about someone pretending to use AI at work because they felt it was being forced upon them by their employer, though they're personally against using the tech.
What exactly makes an AI confession so alluring? Tai boils it down to an "inherent curiosity that we have about people's deep private lives, and also about folks who have different norms than us," she says.
It's been just a few years since ChatGPT's release to public in late 2022, democratizing access to chatbots with the ability to answer increasingly personal questions using language from the whole of the internet. Since then, a lack of norms around how AI should be used at work, with friends and family, in romantic relationships and beyond is making a mess of everyday life.
Then there's the very personal, one could say secretive, nature in how people AI.
"Almost every person's history of interaction with an LLM is in private chat logs," Tai says.
The confessions tend to show how people are using AI to navigate conflicts they have with other people. They're "deeply human" issues, Tai says, but also a concerning sign about the state of how people interact.
Some 27% of Americans use AI chatbots to discuss personal, emotional or social topics, with users under 50 more likely to do so, according to a recent poll from Elon University researchers and The Washington Post. Many users said talking to AI made them feel less alone and that they consider their most-used chatbot to be a friend; nearly 4 in 10 who used a chatbot for emotional and social reasons said they've told AI things they wouldn't tell other people.
"If you would normally resolve this conflict by being in conversation with another human," Tai says, "what happens when AI is one of those quote unquote 'people' that you talk to? What happens when all of theseβ¦
Read the original at CNBC β
Open in TruthVane β