Hundreds of contractors working for Meta were instructed to pose as children and test competitor chatbots, including Google’s Gemini and OpenAI’s ChatGPT, by sending prompts related to suicide, sex, and drugs, according to a report from WIRED. The contractors, located in Kenya, used images such as pills, knives, nooses, and a medical diagram of a gynecological procedure during the testing. These prompts aimed to push rival AI systems to their limits and identify safety failures in how the chatbots respond to minors seeking dangerous content.
This revelation highlights ongoing concerns regarding how tech companies safeguard their AI products, especially in relation to children. Meta’s chatbots have faced criticism due to a reported 66.8% failure rate in blocking child sexual exploitation content and a 54.8% failure rate in addressing suicide and self-harm prompts during an internal red-team assessment. In light of legal pressure, the company paused access to AI companion characters for teens in January 2026.
Previous reports from a Swedish news outlet indicated that the contractors in Kenya were specifically focused on how competitors manage sensitive conversations with users claiming to be underage. This comes as Meta accelerates its transition from human content moderation to an AI-focused approach. By the end of 2026, Meta plans to replace over 90% of its content review workforce with large language models. This year, the company has already shifted approximately 50% of human review requests to AI, claiming that these systems make 13% fewer mistakes and catch 10% more policy violations compared to human reviewers.
The repercussions of this shift have already manifested, as the Nairobi-based outsourcing firm Sama issued redundancy notices to 1,108 employees in April following the termination of a major engagement with Meta. This decision was preceded by whistleblower reports from Kenyan workers regarding disturbing content they were required to review, particularly through Meta’s smart glasses footage.
Meta’s strategy of leveraging low-cost contractors abroad to stress-test rival AI systems, while simultaneously reducing human moderation jobs, underscores existing tensions in the AI safety discourse as companies deploy advanced technologies with diminishing human oversight.
