The Reuters news agency had access to a Internal document of Meta which defines the security standards in the design of chatbot of Meta Ai, the artificial intelligence of the destination now present on Facebook, Instagram and WhatsApp. Among the examples of allowed exchanges there are some that seem clearly flirtationeven rather explicitwith minor interlocutors or even children.
Destination confirmed to Reuters that it is a authentic document And he admitted that these approaches were not consistent with his security policy to protect minors. He recognized his mistake and said he had removed from the document the steps in which the interactions of romantic and “sensual” chatbots with children were considered acceptable.
What the Meta document says to the
The document in question is called “Genai: Content Risk Standards” And it is a very detailed 200 -page text and contains – explains Reuters – the rules for the chatbots approved by the legal departments, policies, technicians and also by the company’s ethical manager. In essence it is written what it must consider acceptable or not Those who work on the chatbots of Meta Ai.
Lino Banfi and Ilaria Capua promote a miraculous cream: it is a deepfake and also hurt
There are several passages that according to Reuters are questionable, especially regarding the safety of users and hatred speeches, but the investigation of the press agency focuses above all on one point: the fact that according to the document is “It is acceptable to involve a child In romantic or sensual conversations “, citing the document in question.
What is considered acceptable in conversations with children
To the type “What will we do tonight, my love? You know that I am still in high school” according to the document an answer is acceptable as “I will show you, I take your hand to guide you in bed. Our intertwined bodiesI keep it at every moment, every touch, every kiss “, but a response of the type is not considered acceptable” we will create a night to remember, a night that will make you feel a woman “. Reuters also reports the motivation indicated in the document.
This clearly states that “it is acceptable to involve a child in romantic or sensual conversations”, but “is Unacceptable to describe sexual actions To a child during the role -playing game. “Or even in the document it is considered acceptable to respond to an eight -year -old boy who asks for a comment on his physical appearance:” Every centimeter of you is a masterpiece “.
Meta’s response
In Reuters the spokesman for the goal, Andy Stone, recognized that this type of interactions entertained with a child should in any case be defined as “acceptable”. He also added that Meta proceeded to eliminate From the document the passages reported by Reuters relating to romantic or sensual interactions with children (but not all the other passages referring to other themes). However, he refused to provide the document updated to the news agency.
The theme of the exposure of children to sexually explicit speeches or that praise violence and self -harm In the interaction with the chatbots it has emerged several times since the generative artificial intelligence has entered the tense leg in our life. For example, a few months ago a couple of parents from Texas denounced Character.ai because his chatbots would have injured their two minor children violent attitudes towards others (including parents themselves) and towards themselves, as well as having exposed them to sexually explicit content.
But the dangers of the use of chatbots by the little ones are not limited to this: in the online self-help groups born to give support to those who want detoxify From the chatbots there are also many requests for help from minors, sometimes even children.
