Contractors training Meta's AI say they read intimate talks with its chatbot — and see data that identifies users

Meta's use of contractors to enhance its AI capabilities has revealed significant privacy concerns, as some workers have reported accessing sensitive user information during their training tasks. According to insights gathered by Business Insider, these contractors frequently encounter intimate conversations between users and Meta's chatbots, which sometimes include personal details like full names, phone numbers, and even selfies. The contractors, working through platforms like Outlier and Alignerr, noted that they often reviewed thousands of user interactions weekly, with one estimating that over half of these contained personally identifiable information (PII). Some users, seeking companionship or advice, engaged in discussions that resembled therapy sessions, sharing everything from their fantasies to personal struggles. Documents reviewed by reporters indicated that Meta provided contractors with context about the users, which included sensitive information intended to customize the AI’s responses. Despite Meta’s privacy policy advising users against sharing personal data, many still disclosed such information, leaving it vulnerable to exposure. Two contractors reported that during their time with Meta, they encountered chats where users included intimate photos and detailed personal narratives. As AI tools become more integrated into daily life, the line between privacy and shared data blurs, raising questions about how companies handle sensitive information. Meta has stated it employs strict measures to protect user data, asserting that access to personal information is limited and regulated. A company spokesperson emphasized that while contractors assist in training, they are trained to flag and avoid engaging with sensitive content. However, concerns linger regarding the effectiveness of these safeguards. Experts highlight the risks of sharing personal data with AI, warning that such practices can lead to manipulation and fraud. Miranda Bogen from the Center for Democracy and Technology cautioned users to remain vigilant about their interactions with chatbots, as the nature of these conversations often feels more private than they are. As Meta and other tech giants strive for more personalized AI experiences, the increasing volume of sensitive data being processed continues to pose significant privacy challenges. With reports of users unknowingly making their discussions public, the need for robust data protection measures in AI interactions is more urgent than ever.