Meta used fake teen accounts to stress-test rival chatbots on sensitive topics

An investigative report by WIRED has revealed that Meta paid hundreds of contractors to pose as teenagers using fake under-18 accounts to probe rival AI chatbots. Operating under the internal code name Cannes, the project directed contractors to send over 45,000 highly sensitive prompts to ChatGPT, Gemini, and Character.AI without their knowledge. The dummy accounts bombarded the competitors' platforms with crisis scenarios involving suicide, self-harm, eating disorders, and drugs to test their safety safeguards.

Meta defended the operation as a routine, industry-standard safety benchmarking exercise aimed at improving age-appropriate digital experiences. However, AI ethics and legal experts have heavily criticized the scale and deceptive nature of the project, warning that it falls into a dangerous gray zone that violates the terms of service of its rivals.

Read the full story on TNW ->