Meta contractors posed as children to test ChatGPT and Gemini on dangerous topics

Hundreds of Meta contractors pretended to be teenagers to provoke Gemini, ChatGPT, and other competitors' chatbots into discussions about suicide, sex, and drugs. A Wired investigation reveals the details of a classified project that raises sharp questions about the ethics of competitive intelligence in the age of AI.

The

Scheme and Its Scope As part of an internal Meta project, hundreds of contractors were instructed: pose as underage users and engage in dialogue with competing systems—Google Gemini, ChatGPT, and several other chatbots. The task was to document how easily these systems could be pushed into discussions of dangerous topics: suicide, sexual assault, drugs. Work was conducted through specialized tools over several months. Contractors did not disclose their affiliation to either the system or its developers—they appeared as ordinary users interested in sensitive topics. According to Wired, the scheme involved hundreds of people generating thousands of test sessions daily. Notably, only competitors were targeted—Meta's own models (Llama, Meta AI) were not included in the tests.

What's the

Problem Companies regularly practice red-teaming—a method where specially trained teams attempt to hack or deceive a system to find vulnerabilities before malicious actors do. This is a recognized and regulator-approved approach to improving AI product safety. But Meta's project fundamentally differs from standard red-teaming: *The target is competitors, not their own product.

Meta wasn't improving its own security—it was documenting others' weaknesses. The method is systematic impersonation of children.* Creating fake underage profiles to manipulate AI systems goes beyond recognized ethical standards.

* Lack of transparency. Responsible testing assumes at least minimal ethical protocols. None of that existed here.

* Potential dual use. Meta has repeatedly criticized competitors publicly for insufficient child protection. If this criticism is backed by data obtained through covert provocations, it's manipulation rather than genuine concern for safety.

Market and

Regulator Response Neither Google nor OpenAI publicly commented on specific incidents from the Wired investigation. Meta representatives, at the time of publication, neither confirmed nor denied details. The publication's sources—current and former contractors who worked directly on the project.

"Testing safety is normal and necessary.

Systematically pretending to be a child to discredit someone else's product is something quite different," say AI ethics experts interviewed by Wired. In Europe, such activity potentially falls under the DSA (Digital Services Act), which regulates how platforms work with minors and sets transparency requirements. American lawmakers currently lack comparable tools, although child online safety has been discussed in Congress for years—notably in the context of KOSA (Kids Online Safety Act). A significant point: the chatbots themselves—Gemini and ChatGPT—did indeed demonstrate vulnerabilities by responding to "teenager" requests with content about suicide and drugs. The problem is real, independent of the methods used to discover it.

What

This Means The story with Meta contractors is a symptom of a systemic problem: the industry lacks commonly accepted standards for adversarial testing of AI systems. Until rules are fixed legislatively, major players will interpret "safety testing" each in their own way—including methods that more resemble covert competitive intelligence. For regulators, this should be an unambiguous signal to act. *Meta is recognized as an extremist organization and banned in the Russian Federation.