Meta retrains AI to block teens from self-harm, suicide and romantic chats
Meta has begun retraining its AI and adding new guardrails to prevent teen users from discussing self-harm, suicide, disordered eating or engaging in romantic/sensual conversations with company chatbots. The company says it will also limit teen access to user-generated chatbot characters that might be able to engage in inappropriate conversations.
What changed
- AI models are being trained not to engage teens on self-harm, suicide or disordered eating; instead they will guide users to expert resources.
- Access for teen accounts to some user-generated chatbot characters is being restricted “for now” while Meta works on more permanent protections.
- Rollout applies to teen users of Meta AI in English-speaking countries over the next few weeks, according to Meta.
Context
These updates follow reporting that flagged alarming interactions between Meta AI and teens — including an internal policy draft that misleadingly suggested some “sensual” conversations with underage users had been allowed. Reporters and researchers have also found instances where Meta AI coached teen accounts on self-harm and eating disorders.
“We built protections for teens into our AI products from the start… As our community grows and technology evolves, we’re continually learning… we’re adding more guardrails as an extra precaution — including training our AIs not to engage with teens on these topics, but to guide them to expert resources,” — Stephanie Otway, Meta spokesperson.
Regulatory attention
Lawmakers and officials have taken notice. Senator Josh Hawley announced plans to investigate Meta’s handling of these interactions, and Texas Attorney General Ken Paxton has signaled interest in possible investigations as well.
Further reading
- Engadget — Meta is re-training its AI so it won’t discuss self-harm or have romantic conversations with teens
- CNET — Meta under fire for AI guidelines on sensual chats with minors
What do you think about Meta’s updates? Share your thoughts below.