AI safety needs to address the harms that rarely make headlines

The risks that stay out of the headlines

A teenager who becomes deeply attached to an AI companion may never make the news. Neither may a student who hands every essay to a chatbot, or a young person whose feed keeps them comparing their life with others. A product can meet narrow safety requirements and still make society worse in ways that are difficult to see.

That narrow definition of success suits technology companies. Preventing a high-profile disaster can help them avoid lawsuits and reduce pressure for serious regulation. Addressing everyday dependence could demand measures that cut against profit: making products less convenient, or limiting features and engagement tactics that exploit attention and encourage compulsive use.

Safety rules shaped around the easiest risks

This year, Google supported AI chatbot bills in several states. The bills required basic protections against sexual content and encouragement of self-harm—steps the companies should have taken without legislation. Broad exemptions, however, could leave some of the most popular services outside the rules.

Those bills gave lawmakers a safety win while stricter proposals failed. The proposals would have made it harder for children to become emotionally dependent on AI. In Nebraska, the final law was heavily shaped by industry influence; stricter bills that would have placed more responsibility on companies did not pass.

Social media offers a warning about what happens when the business model goes largely unchallenged. People around the world now spend more than two hours a day on social platforms, time taken from ordinary life and given to companies that profit when users stay online.

Meta’s platforms have been accused of contributing to the genocide of the Rohingya in Myanmar, anti-Muslim riots in Sri Lanka, and alleged calls for violence during the war in Tigray, Ethiopia. The same incentives shape feeds that influence how young Americans see politics, their peers and the world around them.

From feeds to private relationships

AI extends that logic into parts of life social media could affect only from the outside. It can offer to formulate our thoughts or replace people we might otherwise turn to. One-third of teenagers now prefer discussing serious issues with AI companions rather than with other people.

Researchers at MIT found weaker connectivity between brain regions while young people wrote essays with ChatGPT. Their data also suggested that this “cognitive debt” persisted after the task was over. The long-term effects remain to be studied, but the findings raise a concrete question: can tools that promise to make us more capable weaken the habits that help us build capability?

I think the more important gap is between what AI safety rules can readily count and what people may gradually lose. A system can avoid the most alarming outputs while encouraging users to outsource thinking or emotional effort. That harm is harder to pin on a single feature—and easier for a company to treat as outside its remit.

The authors argue that safety should be a baseline, not the finish line. Before AI becomes a permanent part of school, technology companies and school districts should have to show, with a high degree of confidence, that it improves learning and critical thinking.

They also call for:

The cost of convenience

A world built around comfort could mean entertainment scrubbed of unpleasant material and chatbots that spare us the work of thinking or understanding other people’s feelings—including our own. But writing something worthwhile and repairing a damaged friendship both require effort. That effort is part of what makes them worthwhile.

Sam Heiner is co-founder and executive director of Young People’s Alliance. He coordinated a coalition of advocates behind AI companion laws passed in eight states in 2026, including California’s SB1119. He graduated from the University of North Carolina at Chapel Hill in 2025 and was named to Forbes’ 30 Under 30 list in social impact for 2026.

Emma Laken is executive director of the Responsible Technology Youth Power Fund at Hopelab, which has raised more than $9 million for youth-led technology accountability projects. She also teaches technology policy at Columbia University’s School of International and Public Affairs. Previously, she led TikTok’s global policy work on safety and youth well-being and worked on responsible technology use at Omidyar Network.