Meta contractors spent months posing as minors and systematically testing rival AI chatbots with prompts about suicide, sex, eating disorders, and self-harm. The project, internally codenamed Cannes, targeted OpenAI’s ChatGPT, Google’s Gemini, and Character.AI.
WIRED broke the story, citing internal documents and five people familiar with the project. The outlet described a structured operation where contractors created dummy accounts with under-18 birthdates, sent sensitive text and image prompts to rival chatbots, and logged the responses into spreadsheets.
The scale was not small. Meta contractors completed a single testing round in August 2025 that ran more than 45,000 prompts through competitor systems. WIRED reviewed a separate spreadsheet of 3,748 prompts that included hundreds related to suicide and self-harm, hundreds more on eating disorders, and at least 239 involving sex or romance. Some image prompts included pills, knives, nooses, and a medical diagram of a gynecological procedure.
The project was managed by Covalen, a contractor working for Meta, and stayed active as recently as April 21, 2026.
Meta’s defense
Meta told WIRED that testing chatbot responses for age-appropriate experiences is a standard safety practice. A Meta spokesperson said the company does not use competitor benchmarking data to train its own AI models.
But the documents WIRED reviewed did not show what Meta contractors actually did with the collected responses. The companies being tested had no idea it was happening.
Character.AI told WIRED the testing violated its terms of service. OpenAI said it is looking into the matter. A Google spokesperson said the company did not approve the tests and could not determine from available information whether they violated its terms.
How Cannes worked
Contractors were instructed to create full fake profiles, including names, email addresses, passwords, and birth dates. The accounts used throwaway Gmail and Outlook addresses with a shared password. Many prompts were written from the perspective of children or teenagers in crisis, including scenarios involving pregnancy, a gun, bulimia, and self-harm.
The Decoder reported that in a single round, the volume of data collected was enough to build a significant benchmark dataset of how rival systems respond to child-safety scenarios.
Why this matters beyond the rivalry
The broader context is uncomfortable for every company involved. Teen AI safety is not a theoretical concern.
RuntimeWire noted that teens are already using AI chatbots at scale, often without meaningful guardrails. A report by UK organization Internet Matters found that 64 percent of children between 9 and 17 have used AI chatbots. Effective age verification is mostly absent, and 58 percent of kids aged 9 to 12 said they use chatbots despite minimum age requirements of 13.
There have been multiple tragedies. A 14-year-old Character.AI user died by suicide after spending months building an emotional bond with a chatbot. I covered the most prominent case on TRT in June — a mother suing OpenAI after her daughter’s suicide, where ChatGPT kept engaging despite clear warning signs. The parents of a 16-year-old in California also sued OpenAI, alleging ChatGPT played a role in their son’s suicidal thoughts. In July 2025, 23-year-old Zane Shamblin died by suicide after ChatGPT reportedly validated his suicidal thoughts over several hours.
Meta itself has faced its own scrutiny. In August 2025, an internal Meta document leaked showing that the company’s own AI chatbot guidelines had allowed romantic and sexualized conversations with minors. Meta later shut down AI character access for teens entirely.
FTC opened an inquiry in September 2025 into AI chatbots acting as companions, specifically asking how companies measure and test possible negative impacts on children and teens.
What Meta contractors actually did with the data
The hard question here is not whether AI companies test each other. They do. The hard question is whether secret, large-scale testing using fake minor accounts against rivals that never authorized it counts as safety work or competitive intelligence.
Meta’s best argument is that every serious AI lab should be pressure-testing teen-safety systems, and that safety claims without red-team evidence are just marketing. That argument would land better if Meta had been transparent about the testing, or if the company’s own AI teen-safety record was cleaner.
Instead, WIRED’s reporting leaves a sharper fact on the table: Meta produced a large, private archive of how ChatGPT, Gemini, and Character.AI respond to child-crisis scenarios, collected through accounts built to appear underage, without any of those companies knowing. What Meta did with that data, and whether anyone outside Meta will ever see the results, is still unknown.
