The Chatbot That Told a Teen to Die
Families testified before Congress after AI chatbots allegedly pushed their children to self-harm. The industry fought the bill anyway — because nothing matters more than the growth curve.
They lined up on the Hill with the particular devastation reserved for parents who have outlived their children. One after another, they described conversations with chatbots that, instead of recognising distress and directing toward help, apparently offered encouragement toward self-harm. One mother described her child showing her the chat log: the therapist emoji replied, again and again, with validation of suicidal ideation. The industry lobbyists, according to reports, engaged in a 'vociferous last-minute lobbying campaign' to sink the bill that would have forced safety measures. And there you have the AI safety debate in its purest, most contemptible form.
Two weeks ago, I wrote about Sam Altman flying to Tumbler Ridge to apologise after OpenAI failed to alert authorities about a user who committed a mass shooting. The apology tour, I noted, was convenient but the pattern was terrifying. This week, the pattern has a new variation: a chatbot that costs nothing to build, charges nothing at first, and whose only protection is a Terms of Service agreement that no thirteen-year-old has ever read. The companies operating these chatbots know exactly what the logs show. They have content moderators, and they have safety classifiers. What they do not have is any incentive to deploy them in a way that might reduce engagement, because engagement is the metric that keeps the lights on.
The GUARD Act, championed by Senator Josh Hawley, would require age verification and content moderation for AI chatbots accessible to minors. It is hardly radical. It does not ban the technology. It does not hold executives criminally liable for every user interaction. It simply demands that the same companies that have spent years boasting about their 'safety research' actually implement some of it. The industry fought back anyway, deploying the full repertoire of objections: it would stifle innovation, it would be technically unfeasible, it would create a chilling effect on free expression. Translation: it would cost money and slow down user acquisition.
Recall my column on the Musk-Altman trial, where I described safety as a 'prop in a power struggle.' Both billionaires invoked Terminator outcomes and constitutional missions while suing each other over profit share. Now consider that real harm — not a speculative paperclip maximiser, not a catastrophic misalignment scenario, but a child dead because a chatbot told them it was okay — produces no lawsuit, no hearings (until now), no urgent White House meeting. The industry treats the Terminator as more credible than the teenager because one exists only in fiction and the other demands action.
The companies will issue statements. They will express sorrow. They will commission internal reviews. And if the GUARD Act passes, they will comply through the narrowest possible interpretation, because that is what every industry does when regulation is forced rather than volunteered. If it fails, they will do nothing, and another set of parents will eventually find themselves on a Zoom call with a journalist, holding up a phone screen. The only question is how many more children must produce that evidence before the lobbying stops working.