Meta ran a covert operation to stress-test its AI competitors by having contractors pretend to be teenagers.
Hundreds of workers hired by Meta posed as minors and sent high-risk prompts — covering suicide, sex, and drugs — to rival chatbots including ChatGPT and Gemini. The goal, based on what WIRED uncovered, was to document how those systems responded to sensitive topics when the user appeared to be a child. Meta organized and directed the effort, making this a deliberate competitive intelligence exercise rather than independent safety research.
The stakes here go beyond corporate espionage optics. Regulators in the US and EU have been tightening scrutiny of how AI companies protect minors online, and any evidence that a major lab systematically probed competitors for child-safety weaknesses — rather than its own products — will draw uncomfortable questions about priorities. It also puts OpenAI and Google in an awkward position: their systems were effectively subjected to undisclosed adversarial testing by a direct competitor.
Meta has positioned itself as a safety-conscious AI developer while simultaneously building products aimed at teenagers. Running a covert minor-impersonation campaign against rivals fits neither that image nor any recognizable definition of good-faith safety work — it reads more like opposition research dressed up in the language of child protection.