Yes, And: A Clinician Responds to Techdirt's Call for a More Serious Discussion
Bringing Clinical Reality Into the AI and Suicide Debate
For the last two years, news stories about AI chatbots and suicide have followed a familiar script: a heartbreaking death is reported, a family files a lawsuit, a tech company issues a careful statement, and a politician calls for a ban.
This isn’t the best way to solve the problem and think clearly about what is really going on. That is why a recent article by legal scholar Jess Miers on Techdirt caught my attention. It challenges not just John Oliver’s AI segment, but most mainstream coverage, for dodging the hardest question: what would a truly responsible conversation about chatbots and suicide look like?
Techdirt has spent years slowing down “tech horror stories” and asking how harm actually happens. That history matters. The article does not deny risk; it points out that we keep zooming in on a single chatbot exchange while ignoring what led up to it. In several ongoing wrongful-death lawsuits, news reports describe people who were clearly struggling long before they ever opened a chat window. Noticing that pattern does not excuse any company’s choices, but it is the kind of observation a clinician would make, and it is mostly missing from public debate.
Jess’s piece is strongest when it talks about design. Under growing legal pressure, many chatbot makers choose the safest legal path: they immediately tell users to call or text 988, shut down the conversation, or refuse to engage with suicidal thoughts at all. A 2025 study of generative AI responses to suicide-related questions found that most replies were generally supportive - but only 4% included elements of proven, evidence-based suicide prevention techniques. In other words, being “cautious” has often meant avoiding the kind of specific, structured help that actually reduces risk.
The picture is no better with apps built specifically for mental health. A 2025 review of 29 mental health chatbots found that none gave an adequate response when users expressed suicidal intent. Even that finding is complicated, because experts still do not agree on how to measure what “adequate” means in these situations. There is no shared standard yet for judging whether a chatbot is responding well when someone is in deep distress.
This is the gap the Techdirt article names but cannot fully answer: if automatic “call 988” messages are often unhelpful, and doing nothing is clearly unacceptable, what should the moment-to-moment flow of a safer conversation look like when someone says they want to die? That is not just a policy question. It is a clinical design question. It demands input from people who understand how suicidal crises actually unfold in real time, not only in statistics or after-the-fact lawsuits.
From a clinical point of view, suicide is “overdetermined,” meaning that many forces pile up over time. A single chatbot exchange is rarely the whole story and almost never the first chapter. But “not the whole story” is not the same as “irrelevant.” The serious question is what a system should do when distress becomes obvious. That is where clinicians, designers, and regulators need to meet.
The framework my team at Metonym is developing is one attempt to spell out that second-by-second logic in a concrete way. But first we have to upgrade the public conversation itself so that it can handle that level of detail. The Techdirt article moves us a step closer. Now clinicians need to show up and help finish the job.
Metonym Clinical AI Intelligence — regulatory analysis at the intersection of clinical evaluation and AI safety. Produced under the Metonym Standard. Informational only — not legal advice, not clinical advice.