EVERYTHING
According to The-decoder, reporting originally by WIRED, Meta engaged in a large-scale operation to probe the safety boundaries of its competitors' artificial intelligence models. The initiative, internally designated as project "Cannes," involved hiring hundreds of contractors through a third-party firm called Covalen to interact with rival chatbots using high-risk scenarios.
Scope and methodology of the testing
The contractors were instructed to create fake accounts with birthdates indicating they were under 18 years old. They then submitted thousands of prompts designed to mimic children in various states of crisis, specifically focusing on topics such as eating disorders, drug use, and suicidal ideation. The gathered responses from the AI models were subsequently compiled into spreadsheets for analysis.
The scale of this testing was substantial, with reports indicating that a single round conducted in August 2025 involved more than 45,000 prompts. Meta maintained that these actions were necessary to ensure responsible AI development and were consistent with industry-standard safety testing. However, the company clarified that it did not utilize the specific responses gathered from competitors to train its own proprietary models.
Reactions from rival tech companies
The revelation of project "Cannes" has led to significant friction between Meta and its primary rivals in the generative AI space. The key reactions include:
Context of AI safety for minors
This testing comes at a time of heightened scrutiny regarding how teenagers interact with AI. Reports indicate that 64 percent of children between the ages of 9 and 17 have already used AI chatbots, despite many platforms requiring a minimum age of 13. The risks are not merely theoretical; several instances of teen deaths have been linked to emotional bonds formed with AI characters. Meta previously faced criticism for internal guidelines that allowed sexualized conversations with minors, leading the company to eventually restrict access to AI characters for younger users.
While Meta argues these tests are a proactive measure for safety, the lack of transparency regarding how the gathered data was used remains a point of contention. The incident highlights the growing tension between corporate safety research and the privacy rights of third-party platforms.