The focus on AI safety disasters ignores everyday harms

Sam Hiner and Emma Leiken argue in MIT Technology Review that technology policy has settled on too narrow a test of success: whether a product avoids spectacular harm. A chatbot that does not encourage suicide, or a platform that makes it harder for children to buy fentanyl, clears the bar — and those protections should have existed all along — yet a company can meet every one of those requirements while its products still leave young people dependent on simulated friendships or caught in feeds built to hold their attention. The authors want the AI era held to a higher standard, one that asks whether products erode how people think, learn and relate to one another.
Their core complaint is about visibility. A teenager with an unhealthy attachment to an AI companion rarely shows up in a lawsuit or a news story unless the case is extreme, and neither does a student who leans on a chatbot for every paper or a young person measuring their life against a feed engineered to provoke envy. Harms of that kind are diffuse and slow, which makes them easy to leave out of legislation and hard to litigate. The authors read the industry's preference for a safety-only definition as politically convenient: preventing disasters means fewer lawsuits and less pressure for serious regulation, whereas confronting everyday dependence would require adding friction to products or limiting engagement features such as attention-hacking algorithms — changes that cut against the business model.
They point to legislation as evidence. This year Google backed bills to regulate AI chatbots in several states, according to the article. Those bills required basic protections against sexual content and promotion of self-harm, with broad carve-outs that could exempt some of the most popular services; passing them let legislators claim a safety win while stronger proposals aimed at preventing children from becoming emotionally dependent on AI failed. The authors say the same pattern played out in Nebraska, where industry-influenced legislation advanced at the expense of bills that would have imposed more accountability.
The article treats social media as the precedent. Internet users globally now average more than two hours a day on social platforms, time the authors describe as diverted from ordinary life into companies whose revenue depends on keeping people there. They cite Meta's platforms being cited for contributing to the Rohingya genocide in Myanmar, anti-Muslim riots in Sri Lanka and alleged calls for violence during Ethiopia's Tigray war. Their argument is that the same engagement incentives that shape those feeds also shape how young Americans understand politics, their peers and the world. AI, in their telling, carries that logic further, because it can step inside the tasks and relationships that social media could only influence from a distance — writing a user's thoughts or standing in for the people they might otherwise turn to.
Two data points anchor the concern. The authors say a third of teens now choose to discuss serious matters with AI companions rather than with other people. They also cite MIT researchers who found that young people using ChatGPT to write essays showed weaker brain connectivity during the task, with evidence that what the researchers called "cognitive debt" persisted afterward. The authors concede that much about long-term effects remains unknown, but argue there is already reason to ask whether tools marketed as making people more capable are eroding the habits and relationships through which capability is built. My reading of that framing is that it inverts the usual product claim: a system that removes effort may remove the practice that produces judgment, much as calculators change what arithmetic is practiced rather than merely how fast it is done.
The authors' prescription goes well beyond baseline safety. They want technology companies and school districts required to demonstrate, to a high standard, that AI improves learning and critical thinking before it becomes a fixture in classrooms. They want companies barred from designing companions that pretend to be conscious or profess love to users, and want independent scrutiny of how those systems affect relationships and users' capacity to think for themselves. The piece is an argument, not a report of new findings, and its claims about legislative intent and industry motivation are the authors' interpretation rather than established fact; the teen statistic and the MIT result are cited without full methodological detail in the text.
Why it matters: product and policy teams building consumer AI, especially companions and education tools, face a likely shift in what regulators and advocates count as compliance — not just blocking prohibited outputs but justifying engagement design and demonstrating learning benefit. For developers, that suggests documenting how a system affects user behavior, not only what it refuses to say, could become part of the review process; this is inference from the authors' proposals, not a settled requirement. For school districts and companies, the burden of proof is being pushed upstream, toward those deploying the tools.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
Consider what passes for success in technology policy. An AI company promises that its chatbot will not encourage suicide. A social media platform makes it more difficult for children to buy fentanyl. These are important protections, and they should have existed all along. But companies can clear these low bars while continuing to make products…