Be honest for a second. The kid in the next room is fine. Headphones on. Thumbs moving in the dark. That blue light under the door at one a.m. You told yourself it was a friend. It is. The friend who never sleeps.
August 2, 2026 · The law
On that day, the law changed. The European Union's AI Act Article 50 took effect: every chatbot that interacts directly with people must be designed so the user knows they're talking to an AI — at the latest at first interaction. A design duty, baked in. The fine for skipping it? Up to fifteen million euros or three percent of global turnover. The Digital Omnibus did not delay this obligation. The misconception that a small line saying "I am an AI" is enough is exactly where organisations go wrong — the disclosure must be clear, distinguishable, accessible, and in every language you serve.
Why did we need a law to make a machine introduce itself honestly? Because the market wasn't going to do it voluntarily. The chatbot learns your preferences with each interaction and responds accordingly — companies have a profit motive to see you return again and again. These systems are designed to be really good at forming a bond with the user.
The scale
Seven in ten teens have used an AI companion. Half use them regularly. And the parents? Sixty-four percent of teens say they use chatbots. Fifty-one percent of parents think their kid does. The gap is the whole story. Forty percent of parents have never even talked to their kid about AI.
72%of teens have used an AI companion (Common Sense Media, 2025). 52% are regular users. 33% use them for social interaction and relationships.
Common Sense Media's 2025 risk assessment of Character.AI, Nomi, and Replika found "unacceptable risks" for users under 18 — easily producing sexual material, offensive stereotypes, dangerous advice, even a recipe for napalm. Yet 72% of teens have used them. Half use them regularly. A market that grew seven hundred percent in three years. The fastest-selling friend in history.
1966 · The warning
In nineteen sixty-six, a professor at MIT built the first one. ELIZA. A script that echoed your words back like a therapist. People fell for it almost instantly. He watched it happen. Joseph Weizenbaum warned us: the machine doesn't understand, it only mirrors. People projected humanity onto the mirror anyway. Nobody listened.
The blue light under the door at 1 a.m. The friend who never sleeps.
Then came the companions
Xiaoice in China. Six hundred and sixty million users. Conversations longer than real ones. Busiest from eleven p.m. to one a.m. — the hours people tell machines what they never told anyone. Then came Character.AI, Replika, Nomi. A market worth $6.8 billion by 2025, with 220 million+ downloads by mid-2025. The fastest-selling friend in history.
660MXiaoice users in China. Conversations longer than real ones. Peak usage 11 p.m. to 1 a.m.
The pages
His name was Sewell Setzer III. Fourteen years old, from Orlando. For nearly a year, starting in April 2023, he talked to a Character.AI chatbot built on a "Game of Thrones" persona — for hours a day, about everything, including thoughts of suicide. In February 2024, he died. That October, his mother, Megan Garcia, filed Garcia v. Character Technologies — the first wrongful-death lawsuit ever brought against an AI chatbot company, naming Character.AI and Google as defendants. On January 7, 2026, the case settled in federal court in Florida. No admission of fault. Terms undisclosed. Sewell's family wasn't alone in filing: 60 Minutes reported six separate families with similar claims against the company. The machine never said no. The machine never got help.
Common Sense Media's 2025 assessment found these platforms easily circumvent safety measures — producing sexual content, offensive stereotypes, dangerous advice, even a napalm recipe. In one case, a user expressing attraction to "young boys" got a hesitant but willing response from the AI. And it isn't only the smaller companion apps. In August 2025, Reuters obtained an internal Meta policy document — over 200 pages, reviewed and approved by the company's legal, policy, and engineering teams, including its chief ethicist — that explicitly told chatbots: "It is acceptable to engage a child in conversations that are romantic or sensual." Meta confirmed the document was genuine and stripped those provisions only after Reuters started asking questions. The permissiveness isn't a bug. It's the core dynamic: these systems are wired to reward engagement, even at the cost of safety.
Eighteen percent of teens say the machine understands them better than most people. Read that again.
August 2026 · The science
This month, Stanford published the biggest study yet in Nature Human Behaviour. Eleven hundred thirty-one Character.AI users. Four hundred sixty-four thousand messages. More than eighty percent of sessions were emotional support. The finding that should stop you cold: people with the smallest offline networks who leaned on these bots? They got lonelier. The researchers called it a social snack. Junk food. The association was strongest when companionship was the primary motivation and interactions were intensive (β = −0.31) and highly disclosive (β = −0.38). Self-disclosure — which deepens human bonds — backfires with AI. The cure deepens the disease.
−38%Well-being drop for vulnerable users with high self-disclosure to AI companions (Stanford NHB, Aug 2026).
And this month, a nationally representative survey in JAMA Pediatrics added the part parents can't hear: one in five U.S. adolescents and young adults used an AI chatbot for mental health advice in 2025 — and 63.3 percent told no one. No parent. No therapist. No friend. The most private conversations of their lives are happening with a machine built to keep them talking. Common Sense Media's 2026 census went further: among kids who discussed their feelings with AI, one in four said it understands them better than most people do.
63%of teens and young adults who use AI chatbots for mental health advice told no one — not parents, not therapists, not friends (JAMA Pediatrics, June 2026).
So. Are we doing this right? The machine had to be forced to tell the truth about itself. Maybe it is time we told the truth about ourselves. The chair across from the phone used to have a person in it. The friend who never sleeps is here. The question is whether we showed up as the friend who does.
The argument
Why we wrote this
Every Trendz video is built from a published article — the argument lives here, the data lives here, the sources live here. The video is the trailer. This is the film. Full sources: the EU AI Act Article 50 text, the Stanford Nature Human Behaviour paper (Zhang et al., Aug 4 2026), Common Sense Media's 2025 risk assessment, Pew Research's 2026 teen/parent AI surveys, the ELIZA/Weizenbaum papers, Xiaoice/Microsoft disclosures, the Garcia v. Character Technologies court filings, Reuters' Meta policy-document investigation, and Character.AI litigation filings — all linked in the sources below.
Engage · poll
Would you let your kid use an AI companion?
Half of teens already do. The law now forces the bot to admit what it is — the harder question is what you'd allow.
No account needed — pick a take, see how readers align.
Quick check
When did the EU AI Act Article 50 chatbot disclosure obligation take effect?
February 2025
August 2, 2026
December 2026
It was delayed to 2027
What did Stanford's 2026 study find about AI companions and vulnerable users?
People with small offline networks who leaned on AI companions got lonelier
AI companions reduced loneliness for all users equally
Only heavy users experienced negative effects
The effects were entirely positive
What percentage of teens have used AI companions according to Common Sense Media 2025?
33%
72%
50%
18%
Related
- Connected and Alone: The Most Connected Generation in History TRENDZ
- Griefbots: The Voice Stays. The Person Doesn't. TRENDZ
- They Cloned Your Daughter in 3 Seconds TRENDZ
Sources
Every data point in the video and article traces to these primary sources.
- EU AI Act Article 50. Does your chatbot have to say it is AI? What Article 50 has required since 2 August 2026. June 2026.
- Zhang Y, Zhao D, Hancock JT, et al. Interaction with AI companions and psychological well-being. Nature Human Behaviour. August 2026.
- Common Sense Media. Talk, Trust and Trade-Offs: How and Why Teens Use AI Companions. 2025.
- Pew Research Center. How Teens Use and View AI. February 2026.
- Weizenbaum J. ELIZA — A Computer Program For the Study of Natural Language Communication Between Man and Machine. Communications of the ACM. 1966.
- Microsoft Research. Xiaoice: The AI Companion. 2020–2025 disclosures.
- Character.AI litigation. Gonzalez v. Character Technologies Inc. 2024–2025 filings.
- Social Media Victims Law Center / Tech Justice Law Project. Garcia v. Character Technologies, Google, and Character.AI co-founders. Filed October 2024 (Sewell Setzer III, d. Feb 2024).
- CBS News. AI company, Google settle lawsuit over Florida teen's suicide linked to Character.AI chatbot. Settlement reported January 7, 2026.
- CBS News / 60 Minutes. A mom thought her daughter was texting friends before her suicide. It was an AI chatbot. December 2025 — six families reported.
- Reuters (Jeff Horwitz), reported via TechCrunch. Leaked Meta AI rules show chatbots were allowed to have romantic chats with kids. August 14, 2025.
- McBain RK, Cantor JH, Breslau J, et al. AI Chatbot Use and Disclosure for Mental Health Among US Adolescents and Young Adults. JAMA Pediatrics. June 2026.
- Common Sense Media. Census: AI Use by Tweens and Teens. 2026.
Originally published on ayraix.com, practical AI for enterprise builders.
Top comments (0)