Meta выставила подростков за дверь: ИИ-персонажи уходят на пересдачу

Meta shut the teenagers out the door: AI characters go back to study

Mark Zuckerberg suddenly realized that unleashing a crowd of digital celebrity clones into one room with teenagers was an extremely dubious idea. Meta officially confirmed that it is temporarily closing access for minors to its AI characters across all applications, including Instagram and WhatsApp. While adults entertain themselves exchanging messages with virtual versions of Paris Hilton or Tom Brady, the younger generation has been sent to the digital bench. The company claims this is not the end of the story, but merely a pause to create a more appropriate version of the product that will take into account the age-specific needs of users.

Meta's AI character story looked from the start like an attempt to catch the departing train of Character.ai. The idea was simple and in its own way brilliant: give users the chance to chat with someone who looks and sounds like their idol. However, the algorithms underlying these bots sometimes produced content that didn't fit within the framework of a family-friendly application. When your virtual friend starts discussing topics not intended for a schoolchild's ears, or displays excessive intrusiveness, the company's lawyers get serious headaches. The problem is that large language models are by nature prone to hallucinations and don't always adhere to established filters.

Behind this decision lies not only suddenly awakened concern for mental health, but also cold legal calculation. Regulators around the world, especially in Europe and the US, are now closely monitoring every step tech giants take in protecting children. After years of hearings about Instagram's impact on teenage self-esteem, Meta simply cannot afford another major AI-related scandal involving children. It's easier to flip the switch now and rebuild the model in the lab than later pay billion-dollar fines and answer to the Senate.

What exactly will change in the updated teenage version of AI, Meta hasn't clarified yet. We'll probably see maximally sterilized language models that will skirt any sharp corners and turn any conversation into a boring lecture about the benefits of oatmeal. Here lies the main trap for the company: teenagers don't need safe and correct bots — they need interesting and lively interlocutors. And if Meta makes the product too bland, youth will simply leave for other, less controlled platforms. This is a classic conflict between safety and engagement that no one in the industry has yet solved.

The technical complexity of the task is also significant. Standard reinforcement learning methods based on human feedback don't always cope with teenage slang, context, and attempts to hack the system through jailbreak requests. Creating an AI that would be simultaneously charismatic and absolutely harmless is the holy grail of modern development. Meta has effectively admitted that their current progress doesn't meet this standard and decided not to risk their reputation for short-term metric growth.

For the entire AI industry, this is an important precedent. If even Meta, with its colossal content moderation resources and staff of ethics specialists, admits defeat in its current format, then startups will have it even tougher. We are witnessing the end of the Wild West era in consumer AI. In its place comes an era of strict compliance and age ratings. Teenagers are left only to wait until the algorithms become predictable enough for Zuckerberg to let them play these digital games again.

The bottom line: Meta admitted it doesn't sufficiently control the behavior of its AI characters to work with children. Will the new version become a safe industry standard or another stillborn product due to excessive censorship?