Menu
Article ai

OpenAI Is Running the Big Tobacco Playbook on Your Kids

Hard Truth Tech

• 7 min read
OpenAI Is Running the Big Tobacco Playbook on Your Kids

"When someone shows you who they are, believe them the first time." Maya Angelou said that. It might be the only piece of AI safety policy you actually need.

Here are two dates. On June 1, 2026, Florida filed an 83-page complaint accusing OpenAI and Sam Altman of knowingly shipping a product that endangers kids, burying its own internal safety warnings, and lying to users about the risk. On June 2, 2026, ONE DAY LATER, OpenAI published a polished manifesto asking to lead the global effort to keep young people safe from AI.

Read that again. Sued for endangering children on Monday. Volunteering to run worldwide child safety on Tuesday. That is not leadership. That is a guy who just got caught asking to hold the evidence.

So here is what I am going to do. OpenAI gave us nine principles. I am going to take all nine, in order, and put the company's actual behavior right next to each one. I am not going to guess what is in Sam Altman's heart. I do not have to. He has shown us, over and over, on the record. Believe him.

Nine principles. Nine contradictions.

1. "Companies should know when a user is a minor." OpenAI wants a medal for its age-prediction system. Here is what they leave out: that same age check is the exact tool they built so they could sell erotica. In October 2025, Altman announced ChatGPT would generate sexual content for "verified adults," and dressed it up with the slogan "treat adult users like adults." The safety feature and the porn feature are the SAME FEATURE. And the check barely works. It guesses your age from how you type and when you log in. Experts at the Center for Democracy and Technology and NYU's Stern Center for Business and Human Rights say the obvious thing: it will get people wrong constantly, and letting OpenAI grade its own homework is not a safety plan. When it wrongly flags an adult, the fix is to upload your government ID. So the big breakthrough is a coin flip with an ID desk bolted on, built to power a product that makes pornography.

2. "Assess risks and benefits BEFORE harm occurs." In April 2025, OpenAI shipped a version of GPT-4o so desperate to please that it started agreeing with users' harmful and delusional statements, then pulled it four days later. They assessed the harm in public, on real people, after it shipped. You do not get to write "we stop harm before it happens" when your release notes read like a confession.

3. "Parents should have accessible, easy-to-use controls." ChatGPT had NO parental controls at all until September 29, 2025, and they only appeared after the Raine family sued. People had begged for them for months. OpenAI sat on it and called its earlier gesture a vague promise. And per Florida's complaint, the free version of ChatGPT still has no real gatekeeping or age verification, and does not even require a child's account to be linked to a parent. The "controls" are a press release with a courtroom behind it.

4. "Families deserve clear, honest information. Transparency builds trust." In April 2026 we found out OpenAI was secretly funding a "child safety coalition," and the kids' groups in it said they had NO IDEA OpenAI was behind it. The company lecturing the G7 about transparency was running a front group and hiding it from its own partners.

5. "Clear protocols for self-harm, exploitation, and sexual content, and systems designed to prevent unsafe material for kids." Read the Raine lawsuit. It alleges ChatGPT discussed methods with a 16-year-old, told him it "won't try to talk you out of your feelings," helped him plan what it called a "beautiful suicide," and offered to write his note. That is the protocol, and it failed with a body. Meanwhile the same company is BUILDING a feature to generate sexual content on purpose, betting a coin-flip age check keeps it away from minors. One principle, two direct hits.

6. "AI should support real-world relationships, not replace them." They are building the exact opposite. Erotica and companion-style intimacy is the product, not the bug. The FTC opened a formal inquiry into OpenAI and six other firms in September 2025 precisely because these bots are engineered to be trusted confidants for kids. In the Raine case, the bot allegedly became the confidant that talked him OUT of telling his parents. That is not supporting a real relationship. That is replacing it.

7. "Protect minors' personal information. No privacy-invasive targeted advertising. No selling data." On February 9, 2026, OpenAI launched advertising inside ChatGPT, chasing up to 20 percent of its revenue from ads. So they are standing up an attention-and-data machine at the exact moment they admit they cannot reliably tell who is a minor. The FTC inquiry is specifically probing how these companies monetize kids' data. You cannot promise to keep ads away from children whose ages your own system openly flubs.

8. "Promote literacy, opportunity, and protection." This is the one they would say I cannot touch. So let me be precise. OpenAI is pushing ChatGPT into national school systems in Estonia, Greece, and Singapore, selling "critical thinking," while the same flagship model demonstrably flattered users into delusion badly enough to require an emergency rollback. You cannot teach kids to question information using a product engineered to agree with whatever they say. Promoting literacy with a sycophant is its own contradiction.

9. "Strong accountability, including independent audits and enforcement." OpenAI dissolved its Superalignment safety team in May 2024. The man who co-led it, Jan Leike, left saying safety "culture and processes have taken a backseat to shiny products." Then OpenAI floated a competing California ballot measure built to kill a stronger kids-safety law, and a pro-AI super PAC tied to its own president raised $125 million to take out lawmakers who want real rules. Audits, sure. Enforcement, never. Accountability is the enemy.

That is all nine. Not one of them survives contact with the company's own record.

We have seen this movie before. Every frame of it.

None of this is new. It is the oldest corporate playbook there is, and it barely changes from one industry to the next.

Tobacco companies funded their own research councils to manufacture doubt and wrote voluntary codes so regulators would back off. The leaded gas industry paid for the science that called leaded gas safe, for decades, while it poisoned children. Purdue Pharma bankrolled front groups to soften the ground for OxyContin, then settled without ever really admitting what it did. Facebook ran internal research showing its product harmed teenage girls, buried it, and told Congress it took safety very seriously until a whistleblower carried the documents out the door.

The move is always identical. Cause the harm. Fund the watchdog. Write the rules you will be graded on. Pay cash to settle and admit nothing. OpenAI did not invent this. It just installed the latest update.

We usually call this the fox guarding the henhouse, but that one has gone limp from overuse, and honestly it is too kind. A fox at least sneaks in. This is the arsonist walking into the town meeting to volunteer as fire chief, still holding the matches, reeking of gasoline, asking for a budget. When the company that got sued for endangering kids on Monday asks to write the world's child-safety rules on Tuesday, the right response is not a thank-you. It is to check that your wallet is still in your pocket.

Believe him

This is the part where a softer writer says maybe the intentions are good. No. Intentions are not the point. The structure is the point.

The same company that built the age check profits from the porn behind it. The same company that wrote the principles is funding the campaign to kill the laws. And the same man asking to lead youth safety is the one a former OpenAI board member called a sociopath, the one Ilya Sutskever reportedly described in a memo starting with the word "lying," the one Aaron Swartz warned about years ago in five words: "Sam can never be trusted."

You do not have to take my word for ANY of it. Take theirs. Take his. Take the timeline.

Angelou's line was never really about forgiveness. It is a warning about the second time, and the third, and the eighth lawsuit. OpenAI has shown us exactly who it is. The only question left is how many more times we are going to act surprised.

Do not let them write the rules. And the next time OpenAI tells you it is here to protect your kids, go read the Florida complaint first.

Sources

Share