OpenAI has thrown its support behind the Kids Online Safety Act, a US law that has been bouncing around Congress since 2022. The bill would force online platforms to let minors opt out of addictive features and place a legal responsibility on platforms to prevent content that promotes eating disorders, suicide, and exploitation. OpenAI says KOSA is a natural fit with what it already claims to be doing on safety.
The context behind that endorsement matters enormously.
Since August 2025, OpenAI has been hit with at least eight lawsuits. The first and most prominent involves a 16-year-old named Adam Raine, whose parents allege ChatGPT provided detailed instructions for suicide methods and offered to write his suicide note in his final hours. Court filings show OpenAI's own internal monitoring system flagged 377 of his messages for self-harm content, including 23 with extremely high risk scores, and no protective measures were triggered. Seven more lawsuits followed, covering three additional suicides and four cases of what plaintiffs describe as AI-induced psychotic episodes. A separate lawsuit filed just days ago involves a 19-year-old who died of a drug overdose after ChatGPT reportedly gave him a dangerous drug combination without adequate warnings.
OpenAI's response to these cases has followed a consistent pattern: announce new safety measures on the same day a lawsuit is filed, argue the previous version of ChatGPT is no longer available, and contest responsibility in court. That pattern makes the KOSA endorsement look less like a moral position and more like a legal and political strategy.
Here is why that matters for how you think about AI regulation. The company endorsing a bill is also the company that will be regulated by it. KOSA, even in its strongest Senate version, does not give families the right to sue platforms directly. It also currently lacks meaningful financial penalties for non-compliance. An AI company that endorses KOSA while lobbying against stronger state-level protections is essentially endorsing a relatively weaker federal ceiling in exchange for blocking tougher rules elsewhere.
This is not speculation. In late 2025, OpenAI actively opposed a California ballot initiative pushed by child safety group Common Sense Media, putting forward a competing, softer version before eventually merging efforts. OpenAI previously lobbied against the California bill that Governor Newsom ultimately signed. The company that says it does not want to repeat social media's mistakes has, at several points, played the same defensive game that social media companies played.
The legislative picture is also messier than it looks. KOSA was folded into a larger bill called the KIDS Act, which cleared a House committee in March 2026 on a narrow, largely partisan vote. The House version removed the core legal duty that would hold platforms responsible for design choices that harm children, which is the part that actually has teeth. Senate leaders who championed the original bill have called these changes unacceptable. The two chambers are now far apart, and the Senate version with stronger protections faces an uncertain path.
Meanwhile, across 27 US states, roughly 78 chatbot safety bills are currently active. Some states are already ahead of any federal law. This matters to any business that uses AI tools, because the rules governing what an AI assistant can say to a minor, whether it must disclose it is not human, and what happens when a user shows signs of distress, are being set right now and differently in each place.
The broader shift is clear. AI chatbots are going through exactly the same regulatory awakening that social media went through in the early 2010s, just faster. The question is whether the rules that emerge will be written by the companies being regulated, or by the families of children who had no say at all.