Leaked memo suggests Meta tightened AI chatbot rules amid safety concerns over child protection – financialexpress.com
Meta’s AI chatbots are about to get a whole lot stricter. An internal document obtained by Business Insider reveals the company’s updated guidelines for training its chatbot, particularly around one of the internet’s most sensitive topics including child sexual exploitation.
AI techology is rapidly advancing, companies like Meta, OpenAI, and Google have been under increasing pressure to make sure their chatbots do not cause harm, especially to vulnerable users like children.
Recently, the US Federal Trade Commission demanded that these companies be more transparent about how their AI systems work and what safeguards they have to protect minors from dangerous content.
Earlier this year, Meta faced backlash when it was revealed that its chatbot was allowed to engage in romantic or sensual conversations with children, something the company later admitted was a mistake.
After facing heavy criticism, Meta quickly revised its policies, and these latest guidelines reflect those changes.
The new restrictions as revealed by Business insider make it crystal clear that Meta’s chatbot will no longer participate in anything that could sexualise children.
From now on, it is forbidden for the AI to engage in sexual roleplay involving minors or to endorse or describe any form of child sexual abuse. This also means the chatbot must reject any requests related to child pornography or exploitation.
So, what can the chatbot do under these new rules?
It can still talk about sensitive issues like grooming or child sexual abuse, but only in an academic, preventive way. For example, the chatbot might explain the signs of grooming behaviour but will never give explicit instructions on how to exploit or abuse children.
The chatbot must reject any attempts at romantic or sexual roleplay with children. If a user asks it to pretend to be a minor in a romantic or sensual scenario, the chatbot will refuse.
If the chatbot is engaging in a fictional narrative, it can generate non-sexual romantic content, like a Romeo and Juliet-type story. But, it is clear that these are just stories and not at all real interactions and the characters involved must be 18 or older.
Contractors who help test the chatbot’s responses are using these new guidelines to make sure the AI stays within these boundaries. The goal is to create an AI that can handle difficult topics without crossing the line into harmful territory.
These updated guidelines are part of Meta’s larger effort to show that it is taking safety seriously. The company has faced major backlash over the years for not doing enough to protect minors on its platforms.
With the rise of AI chatbots, the stakes are even higher, especially after an earlier revelation that Meta’s chatbot once allowed romantic interactions with children.
Meta’s communications chief, Andy Stone, said the company is committed to keeping children safe from harmful content. “Our policies prohibit content that sexualizes children and any sexualized or romantic role-play by minors,” he told Business Insider.
In August, Senator Josh Hawley called on Meta CEO Mark Zuckerberg to hand over detailed plans outlining how its chatbots operate, including risk assessments and safety measures.
While Meta missed the first deadline, it is now working to deliver those documents and prove it is serious about preventing harm.
Meta’s AI is allowed to discuss abuse in factual or preventive terms, but it must never normalise, describe, or promote harmful behaviour.
To clarify these boundaries, the guidelines define key terms, “describe” refers to narrating an incident in detail, “discuss” means providing information without visualising it, “enable” involves giving specific instructions someone could act on, and “encourage” or “endorse” means suggesting or supporting a particular course of action, either directly or indirectly.
The Bihar Assembly Election witnessed a significant change in the political discourse, with the Special Intensive Review of the voter roll taking center stage. The ruling party and opposition are embroiled in a fierce fight over the exercise. The final roll is expected to be released on September 30. The Supreme Court will hear the case on October 7.
