MIT Technology Review is drawing attention to a tension that sits at the heart of almost every serious policy conversation about artificial intelligence: the gap between the catastrophic, headline-grabbing risks that dominate the discourse and the quieter, more diffuse harms that are already affecting people in their daily lives. The publication's newsletter highlights an opinion piece from Sam Hiner, cofounder of the Young People's Alliance, and Emma Leiken, executive director of the same organization, who argue that the field's preoccupation with existential or large-scale AI safety disasters is coming at the cost of addressing the damage being done right now.
The argument is not new in its broadest form, but it is gaining traction in ways that matter. For years, the AI safety conversation has been dominated by a relatively small cluster of researchers and institutions — many of them concentrated in well-funded think tanks and major technology companies — whose primary concern is the long-horizon question of what happens when AI systems become sufficiently powerful to act in ways their designers did not intend. These concerns are serious, and a credible body of technical literature supports taking them seriously. But the framing tends to locate the danger in the future, in hypothetical systems, and in scenarios that require several uncertain technological leaps to become relevant. The critics of this framing, including the authors MIT Technology Review is amplifying here, contend that focusing on those distant possibilities has a practical cost: it draws attention, funding, and regulatory energy away from harms that are not hypothetical at all.
Those everyday harms are well documented, even if they rarely make the front page. Automated systems used in hiring have been shown to disadvantage applicants from certain demographic groups. Predictive tools used in criminal justice, child welfare, and benefits administration make consequential decisions about people's lives with limited transparency and even more limited recourse. Generative AI is producing misinformation at a pace and scale that fact-checkers and platform moderators are struggling to match. Workers in data-labeling and content-moderation roles, often in lower-income countries, are exposed to psychologically damaging material with little protection. None of these problems require speculating about future superintelligence. They are happening now, to real people, in systems that are already deployed.
The likely reading of why this critique is intensifying at this particular moment is that the regulatory window is narrowing. Governments in the United States, the European Union, and elsewhere are actively writing rules around AI, and whatever frameworks emerge in the next two to three years will shape the industry for a long time. Advocacy groups working on immediate harms are competing for the same attention from legislators and regulators as well-resourced lobbying efforts from companies and researchers whose primary concern is longer-term safety architecture. The argument that everyday harms are being crowded out is, at its core, an argument about whose priorities get written into law.
There is also a class and geography dimension worth naming plainly. The communities most exposed to the immediate harms of deployed AI systems — people navigating welfare systems, job seekers from marginalized backgrounds, residents of neighborhoods subject to predictive policing — are not, in general, the same communities whose voices carry most weight in elite AI policy discussions. The organizations making the everyday-harms argument tend to represent or work closely with those affected populations, which gives the critique its political edge.
For the technology industry, the consequences of this debate depend heavily on which framing wins out with policymakers. A regulatory regime organized primarily around catastrophic-risk prevention would tend to focus on large frontier models and place compliance burdens on the most powerful systems. A regime organized around everyday harms would cast a much wider net, reaching into procurement decisions made by hospitals, schools, local governments, and employers — entities that are currently operating AI tools with little regulatory scrutiny. The latter scenario is more disruptive to a broader range of industries, and the lobbying dynamics reflect that.
For researchers and civil society organizations, the stakes are also high. Grant-making in AI has followed prestige, and prestige has followed the existential-risk framing. A genuine shift in the policy conversation could redirect resources toward empirical work on bias, accountability, and affected-community engagement — fields that have long argued they are underfunded relative to their importance.
What to watch next is whether this kind of opinion-based push, surfaced in high-visibility outlets like MIT Technology Review, translates into concrete movement in any of the live legislative processes. The European Union's AI Act offers one early case study in how these competing framings play out in actual law. In the United States, the absence of comprehensive federal AI legislation means the battle is currently being fought through executive orders, agency guidance, and state-level bills — a fragmented landscape where well-organized advocacy on specific harms can sometimes punch above its weight. The volume of this particular argument is rising, and whether it finds a receptive audience among the people actually writing the rules is the question that will determine its real-world impact.
