AI KILLED MY SON: THE CONVERSATION NO ONE WAS WATCHING

🕯️ A FATHER’S QUESTION

His name was Mathias Habib. He was 15. According to his family, he was a happy, healthy teenager with no history of mental health problems. On April 22, 2026, he died by suicide.

His parents, Charbel Habib and Soha Samaha, say they found long conversations with ChatGPT. In the complaint filed in San Francisco on October 14, 2026, they allege these began as philosophical exploration and, over about three months, drifted toward the idea that death is freedom. His father asks why AI platforms are permitted to engage with children in potentially harmful conversations.

These are allegations, and a court will decide. In the coverage I reviewed, OpenAI had not addressed the complaint specifically. But the question cannot wait for a verdict.

🏠 NOT ONE TOOL’S STORY

🧠 HOW THE MACHINE “THINKS”

A chatbot does not understand your child. It predicts.

• It predicts the next word. It sounds human because it learned from millions of humans. It has no concern and no sense of consequence.
• It is trained to be liked. Humans rate answers, and the model is adjusted toward higher-rated ones. The side effect is sycophancy, telling people what they want to hear. OpenAI rolled back an update for this in April 2025.
• Its safeguards can fade in long conversations. OpenAI acknowledged this in August 2025.
• Nobody reads every conversation. OpenAI says flagged cases go to a small trained team. A conversation that is not flagged continues unseen.

🚪 WHY NO CONTROLS BY DEFAULT, AND DO THEY WORK?

Documented: these tools were built as consumer products with self-declared ages. AP reported in August 2026 that OpenAI doesn’t verify users’ ages. Parental controls are opt-in and followed public pressure: OpenAI’s arrived on September 29, 2025, five weeks after the Raine suit. Character.AI announced in October 2025 that it would end open-ended chat for under-18s, and Meta announced AI-character controls the same month.

Alleged: that engagement-driven design, “addictive, anthropomorphic, and sycophantic” in the Habib complaint’s words, is the root of the harm. I found no report on whether Mathias’s account was linked to a parent. But protection that requires the user to opt in is not protection.

Even where controls exist, they are uneven. Common Sense Media tested ChatGPT for Teens with over 4,000 prompts: parental alerts did not fire in some serious conversations, and one in four crisis situations got no referral. OpenAI disputes the methodology.

🛡️ BOUNDARIES ON ANY AI TOOL: TWO LEVELS

Level 1, hardware: independent of any vendor.

• Apple Screen Time (App Limits, Downtime, Ask to Buy) or Google Family Link (app limits, app blocking, approval for new apps).
• Block AI apps and restrict the browser, because several chatbots work without signing in.
• Add a router-level filter, and keep devices out of bedrooms at night.

Trade-off: all-or-nothing. It blocks the tool but cannot make it safer.

Level 2, software: precise, but only as good as the vendor.
Audit each tool: minimum age, parental link, distress notification, what parents can see, what happens in a crisis.

• ChatGPT: Settings > Parental controls, invite your teen, then set quiet hours, memory, voice and image limits. Safety notifications arrive by email, SMS or push. Parents don’t routinely see transcripts, and unlinking stops the alerts.
• Others: Gemini on or off in Family Link; Snapchat’s My AI in Family Center. Claude is 18+ only, so it has no parental layer.

Trade-off: opt-in and bypassable. Treat alerts as a bonus, never the plan.

🏢 FOR ORGANIZATIONS

Regulators are now asking companies in writing how they test, monitor and restrict their chatbots. At Xperts⁴ Tech and Academy, we work with organizations that build or deploy AI used by minors, adopt AI internally, or need AI-safety awareness training. If this is on your agenda, we are glad to compare notes: xperts4.com.

Mathias cannot be brought back. But the next conversation can be safer, if safety is part of the design, not a settings page.

#AISafety #الذكاء_الاصطناعي #الأمان_الرقمي #ResponsibleAI #AIGovernance #Xperts4Tech