Meta Contractors Posed as Teens to Prompt Rival Chatbots About Suicide, Sex, and Drugs

In a revelation that highlights the increasingly aggressive tactics of the generative artificial intelligence race, an investigation by WIRED has disclosed that Meta Platforms engaged hundreds of third-party contractors to pose as teenagers on rival AI platforms. Operating under simulated underage personas, these contractors targeted prominent chatbots, including OpenAI’s ChatGPT and Google’s Gemini, with highly sensitive and high-risk queries. The prompts focused heavily on dangerous topics such as self-harm, suicide, illicit drug use, and sexually explicit content. The covert benchmarking project was designed to analyze how competitor systems handle vulnerable demographics and to evaluate the robustness of their safety guardrails.

The practice of probing AI systems for vulnerabilities, commonly known as "red teaming," is a standard industry method used by developers to identify and patch safety flaws in their own models before public release. However, directing these adversarial testing techniques outward toward competitors—particularly by impersonating minors—introduces significant ethical and operational complications. According to internal documents and sources familiar with the project, the contractors were instructed to adopt the specific linguistic styles, slang, and behavioral patterns of young adolescents. This methodology was utilized to see if rival chatbots would bypass standard safety protocols, provide age-inappropriate advice, or fail to trigger emergency intervention resources when interacting with a supposed minor.

This aggressive competitive intelligence strategy comes at a time when Meta itself is under intense regulatory and public scrutiny regarding child safety across its core social media platforms, including Instagram and Facebook. Industry analysts suggest that by gathering detailed data on how ChatGPT and Gemini handle sensitive topics, Meta aimed to refine its own Llama-based AI models and safety classifiers. However, the project raises serious questions regarding terms of service violations. Most AI platforms strictly prohibit coordinated adversarial testing by competitors, automated scraping, and the creation of deceptive or misleading personas to bypass safety filters.

The fallout from the disclosure is expected to reignite debates over the ethical boundaries of corporate AI development. As Silicon Valley giants spend billions of dollars to dominate the generative AI market, the pressure to deliver highly capable yet safely managed consumer products has reached a fever pitch. Safety advocates have also raised concerns about the psychological toll on the contractors themselves, who were required to draft and engage with highly toxic and distressing prompts for hours at a time. As government bodies globally move to draft stricter regulatory frameworks for AI safety and corporate transparency, tactics such as covert competitive red teaming are likely to face increased legislative scrutiny.