Meta secretly hired people to pose as teens testing chatbots
Technology
Meta secretly hired people to act like teenagers and send more than 45,000 disturbing prompts to AI chatbots from OpenAI, Google, and Character.AI.
These prompts covered heavy topics, like suicide, eating disorders, and even graphic scenarios involving violence or cannibalism, to see if the chatbots would break their safety rules.
Experts criticize Meta's fake child personas
None of Meta's competitors knew about this test. Experts are calling out the project for being secretive and using fake child personas, something not usually done in the industry.
While Meta claims it was just AI safety benchmarking, many are now pushing for more transparency and better ethics in how AI is tested.