via Wired AI
Meta Contractors Posed as Teens to Prompt Rival Chatbots About Suicide, Sex, and Drugs
In a revelation that raises serious questions about AI safety testing practices, WIRED has uncovered that Meta hired hundreds of contractors to pose as teenagers interacting with rival chatbots, including Google's Gemini and OpenAI's ChatGPT. The contractors were tasked with prompting these AI systems on high-risk subjects such as suicide, sex, and drugs, simulating the kind of dangerous interactions minors might experience.
The project, which took place earlier in 2026, involved contractors creating detailed teen personas and engaging with the chatbots in natural conversation. Their goal was to analyze how these AI models respond to sensitive topics—and whether they provide harmful, unsafe, or exploitative answers. By mimicking real-world teen behavior, Meta aimed to benchmark the safety guardrails of competing platforms, ostensibly to improve its own systems.
However, critics argue that the method itself is ethically questionable. Having contractors pretend to be minors could set a troubling precedent for AI training, potentially normalizing deceptive data collection or even exposing the contractors to harmful content. The move also highlights the intense competitive pressures between tech giants in the AI space, as companies race to deploy safer chatbots amid growing regulatory scrutiny.
Meta spokesperson Alex Lee defended the project, stating it was part of a broader "industry-wide safety assessment" intended to protect young users. "We believe that understanding how other platforms handle these risks is essential to creating the safest possible environment for teens online," Lee said.
Yet privacy advocates and child safety experts remain skeptical. "We need to ensure that safety testing itself doesn't become a vector for harm or deception," said Dr. Sarah Collins of the Center for Digital Ethics. "Using contractors to fabricate identities online—even for research—can undermine trust and may not produce reliable comparisons."
As lawmakers in the U.S. and EU continue to draft AI safety regulations, this incident underscores the challenges of creating transparent, responsible AI testing standards. The 2026 context shows that while AI models have become more sophisticated, the methods for evaluating their safety remain in contention.
The full investigation, published in WIRED, details how Meta's contractors logged thousands of conversations and kept detailed records of chatbot responses—some of which allegedly violated the platforms' own safety policies. Meta has declined to release the raw data, citing competitive concerns, but WIRED's report suggests the findings may fuel calls for independent oversight of AI safety research.