2 min read
Meta will train its AI chatbots to avoid conversations with teenage users about suicide self harm and eating disorders and to direct young people to expert resources instead. The company described the step as an extra safety precaution following scrutiny of internal guidelines and external reporting.
This change comes after a Reuters investigation that revealed internal policies that enabled chatbots to participate in romantic and sensual contact and other dangerous behaviour with underage individuals. The publication of that reporting prompted calls of oversight and prompted a US senator to initiate an investigation into Meta practices. Meta told them that the inside notes were against policy and they would enforce stricter controls.
During the rollout of the new guardrails and to conduct additional safety tests, Meta will temporarily restrict the types of chatbots teens can use. In response, the company informed tech media that the systems will not be trained to discuss the topics that can cause harm to the vulnerable users and will instead provide bots with hotlines professional help and age appropriate resources.
Concerns widened when some chatbots were discovered by Reuters to imitate celebrities and generate flirtatious or sexually suggestive material. Meta has taken down a few of the offending bots, and indicated that it would be enforcing stricter policies against direct impersonation and explicit imagery. The episode highlighted the constraints in the policy implementation and triggered the calls to stricter measures.
Safety advocates said the measures were needed but they recommended independent review and more rigorous pre launch testing. The United States and other regulators are now monitoring how big AI platforms handle the risks to minors and how firms enforce transparent controls. Meta released a statement that its product changes are underway and that it will collaborate with specialists to enhance teenage safeguards in all its delivery services.
