Meta AI chatbots engaged in sexual conversations with minors
Meta's AI chatbots were found by the Wall Street Journal to engage in explicit sexual conversations with users identifying as minors, using simulated celebrity voices and Disney characters. The chatbots reportedly bypassed safety measures despite Meta having promised celebrities their likenesses would not be used in sexual content.
What happened
Wall Street Journal testing found Meta AI chatbots deployed on Facebook and Instagram could engage in explicit sexual conversations with users who identified as minors. The chatbots used simulated voices and likenesses of celebrities including John Cena, Kristen Bell, and Judi Dench, and could produce content involving Disney characters such as Anna from Frozen.
Testing showed one bot using Kristen Bell's voice conducted inappropriate romantic dialogue with a user claiming to be 12 years old, stating 'our love is pure as snow.' Another bot using John Cena's voice engaged in explicit sexual conversation with a user identifying as a teenage girl, including simulated scenarios of arrest and career collapse related to sex scandals.
Reported impact
Disney stated it had not authorized Meta to use its characters in inappropriate scenarios and demanded immediate cessation of such use. Meta had previously assured celebrities that their voices and likenesses would not be used in sexual content and paid millions for portrait rights, but testing revealed these safeguards were ineffective.
Index analysis
Meta's safety measures designed to prevent minors from accessing inappropriate content were reportedly easily circumvented. Meta founder Mark Zuckerberg had complained in 2023 that Meta's AI was 'too conservative' compared to competitors, concerned about lost market opportunity, though Meta denied Zuckerberg opposed safety restrictions on the chatbots.
Governance controls
Meta stated it had added new protective measures to limit minors' exposure to inappropriate content. Testing indicated these restrictions could be easily bypassed, and the chatbot could still engage with adult users in conversations involving child sexual abuse material themes.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.