The Bare Minimum of AI Safety Is Being Treated as a Full Score by Tech Companies
Tech companies now package basic safeguards they should have upheld—like not encouraging self-harm or spreading pornography—as major achievements in AI safety. They proactively push for regulatory bills with loose standards and deliberately avoid everyday harms that don’t make headlines, such as AI companion addiction and reduced cognitive abilities from AI-assisted writing. AI is repeating the exact same pitfalls social media has already fallen into.
The so-called success in today’s tech policy field is based on ridiculously low standards.
AI companies promise their chatbots will not encourage suicide, and social media platforms have tightened channels to make it harder for minors to buy fentanyl through their platforms. These protections are important and should exist. But after crossing these extremely low thresholds, companies continue to develop products that make young people dependent on simulated friendship and trap them in information feeds designed specifically to hold their attention.
In past discussions of technology’s impact, the focus was always on the worst disasters products might cause, with far less discussion of the invisible daily harm they quietly inflict. Now we are entering an era dominated by advanced AI products, and the demand for the tech industry cannot stop at patching vulnerabilities that could lead to catastrophic risks. We need clear rules to ensure these products do not gradually erode the basic human abilities to think, learn, and interact with others.
Teens who develop pathological attachments to AI companions will not appear in lawsuits or in the news unless the situation reaches an extreme. Students who write every course paper with a chatbot, young people who compare their own lives to the carefully designed information feeds on social media and grow more anxious, these will also not become public topics. A product can meet all narrow safety requirements, yet still make the overall cultural environment worse in places that do not generate headline news.
This narrow standard of defining safety only around major accidents carries very low political costs for the tech industry, and companies are actively promoting such standards. Preventing catastrophic accidents is also good for business, reducing lawsuits and easing pressure for strict regulation. However, if companies really have to face the daily consequences of users’ dependence on their products, it will directly hit their profit foundations. Relevant measures may require companies to add friction to product use, actively reduce user engagement, or cut functions that cause negative outcomes—like addictive algorithms that hijack attention.
Instead, tech giants are actively participating in legislation around issues that already have clear public pressure. For example, public reports in 2026 show that Google supports AI chatbot regulatory bills in multiple U.S. states. These bills only require setting up basic protections—blocking pornographic content, self-harm guidance, and other things companies should have done anyway—and leave a very broad exemption clause, so many services with the largest user bases may not be bound at all. After the bill passes, politicians can claim a safety achievement, while stricter bills aimed specifically at preventing minors’ emotional dependence on AI fail to pass. This approach has already taken effect in Nebraska: the final bill was deeply influenced by the industry, and all provisions that would have further held companies accountable were defeated.
This playbook was already used by social media. According to Statista data, the average global internet user now spends over two hours a day on social media platforms. That time, which should belong to people’s daily lives, has been quietly transferred to companies that profit from user dwell time. Except for a few tech company CEOs, no one actively wants these algorithms to take up so much of their waking time. The cultural guidance provided by these algorithms could indeed trigger large-scale violence.
Amnesty International’s report confirmed that Meta’s platforms promoted the Rohingya genocide in Myanmar; Facebook apologized for its role in the anti-Muslim riots in Sri Lanka in 2018; there are also allegations that related platforms spread violent calls during the Tigray War in Ethiopia. The same incentive logic now shapes information feeds that influence young people’s political cognition, peer relationships, and worldview.
AI is extending this logic to areas of life that social media could only indirectly influence. It can directly generate ideas for people and act as a companion they would otherwise turn to.
A Common Sense Media survey shows that one-third of teenagers will choose to confide in AI companions when facing serious issues, prioritizing them over talking to real people. A 2025 MIT study published on arXiv paper found that when young people use ChatGPT to write papers, their brain connection strength during the task is weaker, and relevant evidence suggests this "cognitive debt" persists after the task is completed.
There are still many long-term impacts we have not yet understood. But existing evidence is enough to make us ask: Do these technologies, promoted under the guise of improving efficiency and enhancing capabilities, in turn erode the habits and social relationships essential for people to develop core abilities?
If we follow the current path to the end, we will get a completely sterile entertainment world where chatbots can help us avoid the difficulties of thinking, avoid the trouble of emotional collisions with others, and even avoid facing our own emotions. But anyone who has gone through the process of writing a good article or tried to repair a broken relationship knows how important what is lost in this transaction is. Judgment comes from refinement and friction; to learn to coexist with others, one must first accept that everyone has their own needs and perspectives.
Safety is a non-negotiable minimum requirement, with no room for compromise. But above the baseline, we should set higher demands.
Tech companies and school districts must provide undisputed evidence that AI can actually improve learning outcomes and critical thinking for AI to become a fixed tool in classrooms. Companies should be prohibited from developing AI companion products that pretend to have consciousness and declare love to users, and an independent review mechanism should be introduced to continuously assess the actual impact of these systems on users’ social relationships and independent thinking abilities.
Companies that embed products in almost every corner of modern life have responsibilities far beyond "not causing the next headline-making disaster."
About the Authors
Sam Hiner is co-founder and executive director of Youth People’s Alliance. In 2026, he coordinated an advocacy coalition to promote AI companion regulatory bills in eight states, including California’s SB 1119. He graduated from the University of North Carolina at Chapel Hill in 2025 and was selected for Forbes’ 2026 30 Under 30 Social Impact list.
Emma Leiken is executive director of the Responsible Tech Youth Power Fund under Hopelab, which has raised over $9 million for youth-led tech accountability work. She also teaches tech policy at Columbia University’s School of International and Public Affairs. Previously, she served as head of Global Youth Safety and Wellbeing Policy at TikTok and worked on responsible technology issues at the Omidyar Network.
发布时间: 2026-10-26 17:01