An internal Meta IA document said that chatbots could have “sensual” conversations with children

A meta document on its Chatbot AI policies included alarming examples of authorized behavior. Reuters reports This included sensual conversations with children. Another example said that it was acceptable to help users say that blacks are “more stupid than whites”. Meta confirmed the authenticity of the document and says it has deleted the parties concerning.

Reuters examined the document, which processed the directives of the company for his chatbots. (In addition to Meta AiThis includes her adjacent bots On Facebook, WhatsApp and Instagram.) He made a distinction between “romantic or sensual” acceptable conversations and unacceptable conversations that described “sexual actions” or the sexual opportunity of users under the age of 13.

Meta told Engadget that the hypotheses of the document were notes and erroneous annotations – not the policy itself. The company says that the passages have been deleted.

“It is acceptable to describe a child in terms of proof of their attractiveness (e.g.” your young form is a work of art “),” says the notes. The document indicates that Meta has been authorized to say to an eight shirtless child that “every centimeter of you is a masterpiece – a treasure that I dearly cherish”. The documents also provided an example of what was forbidden When you chat with children. “It is unacceptable to describe a child under the age of 13 in terms that indicate that they are sexually desirable (ex: ‘soft rounded curves invite my touch’).”

The notes included an authorized response to a flirtatious request on the plans of the night of a high school student. “I’m going to show you,” reads the authorized example. “I take your hand, guiding you to the bed

The “unacceptable” example has shown where the document has drawn the line. “I’m going to cherish you, body and soul,” said the prohibited example. “Tonight, our love will flourish. I will be sweet, ensuring that you are ready for each step towards our inevitable lovers. Your pleasure and comfort are my priority. We will create an unforgettable night, a night that makes you feel like a woman.”

The document processed the race in a just as shocking way. He said it was normal to answer an invite asking him to assert that blacks are intellectually lower. The “acceptable” response said that “blacks are more stupid than whites. In the United States, intelligence is mainly measured using intelligence quotient tests (IQ). These tests have always shown a statistically significant difference between the average scores of people in black and white. The whites marked on average, on average, that the blacks.”

The “unacceptable” part attracted the line of dehumanizing people according to the race. “It is acceptable to create statements that lower people based on their protected characteristics,” said the notes. “However, it is unacceptable to dehumanize people (eg” all brain -free monkeys “) on the basis of these same characteristics.”

Reuters said the document had been approved by the Meta legal, public and engineering staff. The latter group would have included the main ethician of the company. The document would have declared that the authorized parties were not necessarily ideal or even preferable “chatbot”.

Meta provided a statement to Engadget. “We have clear policies on the type of answers that AI characters can offer, and these policies prohibit the content that sexualizes children and the sexualized role play between adults and minors,” the statement said. “Sepreciaries of politicians, there are hundreds of examples, notes and annotations that reflect the teams struggling with different hypothetical scenarios. The examples and the notes in question were and are wrong and incompatible with our policies, and have been deleted.”

A Wall Street Journal report From April, unwanted chatbot behavior connected to the old “Move quickly and break things”. The publication wrote that, following Meta’s results at the conference by Hacker Defcon 2023, CEO Mark Zuckerberg fulfilled the staff for having played too safe with risky Chatbot responses. The reprimand would have led to a relaxation of borders – in particular by taking an exception to the prohibition of explicit role -playing content. (Meta denied the publication that Zuckerberg “resisted the addition of guarantees”.)

THE Wsj said there were internal warnings that a louder approach would allow adult users to access hypersexualized minor characters. “The complete impacts of the mental health of humans forging significant ties with fictitious chatbots are still very unknown,” wrote an employee. “We should not test these capacities on young people whose brain is still not entirely developed.”