In a federal courthouse in Santa Fe, on the afternoon of 24 March 2026, twelve New Mexicans did something that no jury in the United States had ever done. After a six-week trial, they returned a verdict finding that Meta Platforms, the company that owns Facebook and Instagram, had built products that harmed children, that the company's own executives understood this, and that they had deployed those products anyway. The penalty was 375 million dollars, calculated at the statutory maximum of 5,000 dollars per violation under New Mexico's Unfair Practices Act, multiplied across tens of thousands of breaches. The number itself was almost beside the point. What mattered was the finding underneath it: a body of ordinary citizens had looked at the internal machinery of a social media company and concluded that the harm was not an accident. It was the design.
The verdict landed in a strange and revealing month. In the same weeks that the Santa Fe jury was deliberating, two other documents were circulating that, taken together with the trial, sketch the outline of a problem the law has only just begun to name. One was a clinical analysis in Psychiatric Times describing what its authors called an empathy crisis. The other was a piece of reporting from Outlook India, dated 11 March 2026, about psychologists who had begun to describe something they were observing in their consulting rooms and in the wider population as compassion fatigue operating at the scale of an entire society. Read in isolation, each is a story about a different thing: a lawsuit, a neurological phenomenon, a cultural mood. Read together, they describe a single mechanism and its consequences, and they raise a question that is at once technical, clinical, legal, and moral. If the systems through which hundreds of millions of people now experience the world are measurably eroding their ability to feel for one another, what would it mean to build those systems differently, and who, exactly, has the standing to demand it?
The Mechanics of the Feed
To understand why the question is so difficult to dismiss, you have to start with the engineering, because the engineering is where the harm is located.
A recommendation algorithm is, at its core, a prediction engine. Its job is to guess which piece of content, from among the effectively infinite supply available, will keep a given user engaged for the longest possible time. The system does not have opinions. It does not know what cruelty is, or grief, or war. It has a metric, usually some composite of watch time, clicks, shares, comments, and re-engagement, and it relentlessly optimises for that metric by serving up whatever the data suggests will move it upward. This is not a caricature of how these systems work. It is a description of their explicit objective function.
The trouble begins with what the data reveals about human attention. Content that provokes strong negative affect, outrage, fear, disgust, the sight of suffering, tends to generate more engagement than content that provokes calm or contentment. This is not a flaw in the algorithm; it is a feature of the species. Our nervous systems evolved to prioritise threat. We attend to the snarling face in the crowd before we notice the smiling ones, and a feed that learns this will, with perfect mechanical indifference, escalate. It will serve a user a progressively more extreme diet, because the extreme is what holds the gaze.
The Psychiatric Times analysis published in early 2026, titled in part around how social media algorithms drive emotional numbing, frames this with clinical precision. Desensitisation to violent, high-arousal content, its authors argue, is now a measurable phenomenon, one that reshapes how people experience empathy, form moral judgements, and understand the suffering of others. The crucial claim, and the one that turns a familiar complaint about social media into a genuine indictment, is that the amplification of violent and distressing content is not an unintended side effect of optimising for engagement. It is a predictable consequence of optimising for engagement without any constraint on emotional valence. If you build a machine to maximise attention, and you place no governor on the emotional cost of the content it surfaces, the machine will discover, on its own, that human distress is reliable fuel. The harm is structurally embedded in the design rather than incidentally produced by it.
This distinction is everything. An unintended side effect can be patched. A structural consequence has to be designed out, and designing it out means accepting a lower number in the column the entire business is built to maximise.
A Skill, Not a Trait
The reason any of this should alarm us, rather than merely irritate us, rests on a body of research from developmental and social psychology that has been accumulating for decades and that runs counter to a comfortable intuition. The intuition is that empathy is a fixed quantity, something you either have or lack, a fact of temperament settled at birth. The research says otherwise.
Jamil Zaki, a psychologist at Stanford University and the author of a 2019 book on the subject, has spent years assembling the evidence that empathy behaves less like an inborn trait and more like a trainable capacity. “Empathy isn't doled out to us in a fixed quantity at birth,” he has written. “It's a skill that improves each time we use it.” His preferred metaphor is muscular. “Empathy is something like a muscle: left unused, it atrophies; put to work, it grows.” Through deliberate practice, he argues, through compassion meditation, diverse friendships, even the reading of fiction, people can cultivate their capacity to feel with others. And, strikingly, the mere belief that empathy is a skill rather than a fixed quantity inspires people to try harder at it, and to succeed.
The flip side of a trainable capacity is a degradable one. If empathy grows with use, it shrinks with disuse, and it can be actively damaged by sustained exposure to conditions that reward emotional shutdown. This is not speculation. It is one of the more robust findings in the literature on media violence. In a 2009 study with the deliberately chilling title “Comfortably Numb: Desensitizing Effects of Violent Media on Helping Others,” the researchers Brad Bushman and Craig Anderson demonstrated that people exposed to violent media were measurably slower to help an injured stranger, and reported the emergency as less serious, than people who had not been so exposed. Their conclusion was that exposure to violence in media leaves people, in their phrase, “comfortably numb” to the pain and suffering of others.
The numbness is not merely behavioural. It registers in the brain. Neuroimaging work published in the journal Social Cognitive and Affective Neuroscience, under the equally evocative heading “Emotionally anesthetized,” found that even brief exposure to film violence produced changes in the neural signatures associated with processing emotional faces. Participants who had just watched a violent film showed altered electrical responses to images of happy and fearful faces, a pattern the researchers interpreted as desensitisation: the emotional content was registering with less force, and required fewer cognitive resources to manage. The machinery of feeling was, quite literally, turning its volume down.
Put the two bodies of evidence side by side and the implication is hard to escape. Empathy is a skill that atrophies without practice and is damaged by repeated exposure to distress that one is powerless to act upon. Recommendation algorithms, optimised for engagement without constraint on emotional valence, deliver precisely that exposure, in unlimited supply, to hundreds of millions of people, several hours a day. The systems are, in effect, running a vast uncontrolled experiment in the de-conditioning of human compassion.
The Crisis at Population Scale
What does that experiment look like from the inside of a clinic? This is the territory the Outlook India reporting maps, and it is worth dwelling on because it moves the discussion from the laboratory to the lived.
The article, published in March 2026 under a headline drawn from the words of a distressed person, “I Am Not Well,” set out to document how constant exposure to war and atrocity, mediated through screens, was reshaping the way ordinary people processed suffering. It gathered the observations of clinicians and scholars who deal with this professionally, and their testimony is consistent and unsettling.
Zoya Mir, a clinical psychologist who has worked with people affected by the long conflict in Kashmir, offered perhaps the most precise framing of the mechanism. “It is not that empathy disappears,” she observed, “but when the brain is repeatedly exposed to trauma it begins to dull emotional intensity as a way of protecting itself.” This is a vital point, and it complicates any simple moralising about a hard-hearted public. The numbing is not a character flaw. It is a defence mechanism, the mind throwing up a wall because the alternative, feeling the full weight of every catastrophe scrolling past, is unsustainable.
Yaqeen Sikandar, a Turkey-based psychologist specialising in trauma and cognitive behavioural techniques, described the cognitive sleight of hand by which the mind manages the unmanageable. “When the scale of loss becomes too large to emotionally process,” Sikandar said, “the brain turns tragedy into something countable.” A death becomes a number; a massacre becomes a statistic; the human reality is filed away into a form the mind can hold without breaking. Psychologists have a name for this phenomenon, psychic numbing, and a related one, compassion fade, describing the counter-intuitive finding that our concern can actually diminish as the number of victims rises.
Sanjeev Jain, a senior psychiatrist at the National Institute of Mental Health and Neuro Sciences in Bengaluru, situated all of this in a longer historical frame. The idea that societies can grow numb to large-scale violence, he noted, is not new. Societies have always been capable of growing indifferent to mass death when, as he put it, killing becomes detached from its human context, “just another procedure to be followed.” What is new is the delivery mechanism. The numbing that once required the bureaucratic distance of a war machine or the propaganda of a state can now be manufactured, at retail, by a feed that has learned what holds attention and serves it without pause.
The digital-rights expert Apar Gupta, also quoted in the Outlook India piece, drew the connection explicitly, noting that social media algorithms amplify emotionally charged content and so contribute to the compassion fade that leaves users emotionally exhausted and disengaged. Here the two stories, the clinical and the technological, meet. The algorithm is not merely a neutral pipe through which the world's suffering happens to flow. It is an active selector, an amplifier, choosing the most arousing material precisely because arousal is what it has been built to harvest.
The result, the clinicians suggest, is a kind of mass-scale compassion fatigue: not the burnout of a single overworked nurse or aid worker, the context in which the term was first developed, but a diffuse, population-level dulling of the capacity to respond to distress. And it does not make people better informed. That is the cruel irony at the heart of the matter. The constant exposure does not produce a more engaged, more compassionate, more globally conscious citizenry. It produces the opposite: people who have seen so much suffering that they have stopped being able to feel it, who are saturated rather than mobilised, anaesthetised rather than activated.
What the Jury Saw
This is the context into which the New Mexico verdict must be placed, because the trial was, in a sense, the first time a legal system was asked to render judgement on the design itself rather than on any single piece of content.
The case was brought by the New Mexico Department of Justice, which had filed suit alleging that Meta's platforms exposed children to harm and that the company had misrepresented their safety. Over six weeks, the jury heard testimony that went to the heart of the company's knowledge and intent. Witnesses described internal warnings disregarded, public assurances that diverged from private understanding, and design choices that prioritised engagement over the welfare of young users. Representatives of law enforcement and of the National Center for Missing and Exploited Children testified that Meta's reporting of crimes against children on its apps had been deficient, with the company, in the testimony, generating high volumes of low-quality reports by over-relying on automated moderation.
The jury found Meta liable on the claims brought under the state's Unfair Practices Act. It concluded that the company had concealed what it knew about the dangers of child sexual exploitation on its platforms, that it had made false or misleading statements about the safety of those platforms, and that it had engaged in trade practices the jury was willing to label “unconscionable,” unfairly exploiting the vulnerabilities and inexperience of children. The prosecuting attorney Donald Migliori captured the theory of the case in a single sentence, describing Meta as “choosing to engineer its algorithms to keep young people online while knowing that children are at risk.”
That word, “choosing,” is the hinge. The verdict did not rest on the claim that bad things sometimes happen on Meta's platforms, which no one disputes and which would not, on its own, establish liability. It rested on the finding that the company had made deliberate engineering choices to maximise engagement, that it understood the harms those choices produced, and that it proceeded regardless. Fortune, reporting on the verdict the following day under a headline confirming the finding in the plainest terms, summarised the conclusion that Mark Zuckerberg's social media products were harmful for children and that the jury believed the company knew it.
Meta, for its part, said it respectfully disagreed with the verdict and would appeal, maintaining that it works hard to keep people safe and is candid about the difficulty of identifying and removing harmful content and bad actors. The appeal is a near certainty, and the legal fight is far from settled.
But the significance of the verdict does not depend on whether the specific 375-million-dollar penalty survives. New Mexico's case was selected from a consolidated group of hundreds of similar lawsuits precisely to test whether a product-design theory of liability, the argument that the harm lives in the architecture rather than in any individual post, could survive a jury trial. It could. Within the same period, a separate jury found Meta and Google's YouTube negligent in a related social media harms trial, and analysts noted that a wave of further litigation, including cases brought by hundreds of school districts under public-nuisance theories, was advancing through the courts. If the damages established in these early cases are even partially scaled across the thousands of pending claims, the financial exposure for the platforms moves from hundreds of millions into the billions. The bellwether has rung. For every plaintiff's attorney and state attorney general watching, the message is that this legal pathway is now viable. For the platforms, the message is that the courtroom is no longer a safe harbour.
The Engagement Trap
Why, then, do the platforms not simply build the better system? Why persist with an architecture that produces harms a jury is now willing to call knowing and deliberate? The answer is not that the engineers are cruel. It is that the incentive structure is, in a precise sense, captured.
A social media company makes money by selling attention to advertisers. Attention is the inventory. Every additional minute a user spends in the feed is another increment of inventory to sell, another opportunity to learn more about that user and so to target them more precisely. The recommendation algorithm exists to manufacture that attention, and it is judged, internally, by how much of it it can produce. An engineer who proposes to alter the algorithm in a way that reduces engagement, even for the most defensible reasons, is proposing to reduce the company's revenue. That is a difficult conversation to win, because the metric that would improve, some measure of user emotional health, is diffuse, hard to quantify, and shows up nowhere on the balance sheet, while the metric that would suffer, engagement, is measured to four decimal places and is reported to investors every quarter.
This is what it means to say the harm is structural. It is not that anyone in the building wishes to numb the public's capacity for compassion. It is that the system has an objective function, and the objective function does not include compassion, and so the system optimises it away as surely as water finds the lowest point. The Psychiatric Times analysis makes this case in clinical language, but the logic is the logic of any optimisation process: what you do not constrain, you sacrifice.
It follows that exhortation will not fix this. Asking the platforms to be more responsible, to consider the wellbeing of their users, to think about the long-term consequences, is asking them to act against the gradient of their own incentives, which is to say, asking them to do something companies almost never do absent external pressure. The history of every industry that has produced diffuse public harms while generating private profit, from tobacco to leaded petrol to industrial pollution, suggests that the gradient does not bend on its own. It is bent, if at all, by some combination of liability, regulation, and the slow shift of public expectation. Which is precisely why the New Mexico verdict matters beyond its dollar figure: it introduces, for the first time at real scale, a cost on the other side of the ledger.
Designing for the Feeling, Not the Click
So what would it actually mean to redesign these systems for human emotional health rather than for engagement? The question is not rhetorical, and there are concrete, if difficult, answers.
At the level of the individual clinician, the Psychiatric Times authors propose a set of interventions that are modest but real. They suggest that clinicians be given standardised guidance on how to screen for and discuss a patient's algorithmic exposure during routine appointments, treating the feed as a health variable in the way that diet or sleep already are. They recommend behavioural strategies that patients can adopt: limiting use of these platforms when tired or emotionally dysregulated, when defences are lowest; creating defined viewing windows rather than scrolling intermittently across the whole of the waking day; and, intriguingly, deliberately engaging with neutral or prosocial content so that the recommendation system itself recalibrates, learning to serve a gentler diet. There is something almost subversive in that last suggestion: a recognition that the user can, with effort, partially retrain the very machine that is training them.
But these are coping strategies, and the authors are candid that the long-term solution lies elsewhere, in changes to how the platforms design and regulate their recommendation systems in the first place. This is where the harder, more consequential redesign would have to happen, and it is possible to sketch its outlines.
A system designed for emotional health rather than raw engagement would, at minimum, place a constraint on emotional valence into its objective function, the very constraint the current systems conspicuously lack. It would treat the relentless escalation toward more extreme content not as a success to be reinforced but as a failure mode to be detected and damped. It would build in friction, the deliberate introduction of pauses, of off-ramps, of moments that interrupt the frictionless scroll that the clinicians identify as a driver of compassion fade. It might weight its recommendations toward content that leaves users feeling more capable of action rather than more saturated with helpless distress, distinguishing, in effect, between exposure that informs and exposure that merely numbs. None of this is technically impossible. The same machine-learning sophistication that can predict, with eerie accuracy, which video will hold a user for another forty seconds could in principle be turned toward predicting which content leaves them better rather than worse. The obstacle is not capability. The obstacle is that the better system makes less money, and nothing in the current structure rewards building it.
There is also a deeper design question lurking here, one that goes beyond tweaking a model's weights. The current architecture treats the human being as a quantity of attention to be maximised, an extraction target. A genuinely different architecture would have to treat the human being as an end rather than a resource, which is less an engineering problem than a reorientation of purpose. That reorientation is unlikely to come from within a company whose every incentive points the other way. It is more likely to be imposed from without.
Who Has the Standing to Demand It?
Which returns us, finally, to the question of standing, in both its legal and its broader sense. If these systems are degrading a capacity that developmental psychology tells us is essential to moral life, and if a jury has now found the companies building them liable for knowing harm, who is positioned to demand that they be built differently?
The New Mexico verdict supplies one answer, and it is a significant one. State attorneys general, suing under consumer-protection statutes, have demonstrated that they can establish in court the very thing the platforms have long denied: that the harm is real, that the company knew, and that the design was a choice. This is standing in the most literal legal sense, and the wave of consolidated litigation now moving through the courts suggests that this answer will be tested many more times in the coming years. Litigation is a blunt and slow instrument, but it has one decisive virtue: it forces the production of evidence under oath, dragging into the daylight the internal knowledge that public-relations statements are designed to obscure.
Legislators and regulators supply a second answer, and a potentially more comprehensive one, because they can address the structure rather than the individual case. A liability regime decides who pays after the harm is done; a regulatory regime can, in principle, require that the harm not be done in the first place, by mandating constraints on the optimisation itself, by requiring transparency into how the systems work, by establishing standards for what a feed served to a child may and may not do. The political will for this has historically lagged the evidence, but the evidence, after a verdict like New Mexico's, is harder to ignore.
Clinicians supply a third kind of standing, quieter but indispensable. They are the ones who see the consequences arrive in the consulting room, who can document the desensitisation and the irritability and the dulled capacity to care, and who can translate a population-level phenomenon into the specific, undeniable language of a patient who is not well. Their authority is epistemic rather than coercive, but it is the authority on which any eventual reckoning will rest, because it supplies the proof that the harm is not a metaphor.
And then there is the rest of us, the hundreds of millions whose attention is the inventory, whose nervous systems are the experiment, and whose capacity for compassion is the thing being slowly drawn down. Our standing is the most diffuse and the hardest to organise, but it is also, in the end, the foundational one, because the entire edifice of attention extraction rests on our continued participation. The clinicians' advice about viewing windows and deliberate prosocial engagement is, at bottom, a recognition that the muscle Jamil Zaki describes can be exercised as well as atrophied, that the same plasticity that allows empathy to be eroded allows it to be rebuilt, and that this is not entirely outside individual control.
But it would be a mistake, and a convenient one for the platforms, to leave the whole weight of the problem on individual willpower. The asymmetry is too great. On one side stands a single person trying to ration their own scrolling; on the other, some of the most sophisticated predictive systems ever built, designed by thousands of engineers and refined on the behaviour of billions, dedicated to the single purpose of capturing precisely the attention that person is trying to withhold. To tell that person simply to try harder is to misunderstand the contest. The New Mexico jury, in its way, understood this. It declined to treat the harm as the fault of the children who used the products and located it instead in the design of the products and the choices of the people who built them.
The Wager on Empathy
There is a temptation, in writing about all of this, to reach for despair, to conclude that the machine has won, that the numbing is irreversible, that a generation raised inside the feed will simply feel less than its predecessors and that nothing can be done. The evidence does not actually support that conclusion, and the despair, conveniently, serves the interests of those who would prefer that nothing change.
The single most important finding in the developmental literature, the finding that runs through Zaki's work and through the clinical observations gathered in the Outlook India reporting, is that empathy is not fixed. It is a capacity, a skill, a muscle. It can be damaged, yes, by exactly the conditions the recommendation algorithms create, the relentless exposure to suffering one is powerless to relieve, the reward for emotional shutdown, the dulling of intensity as a defence. But a capacity that can be damaged can also be restored, and a system that was designed one way can be designed another. The plasticity cuts both ways. That is the whole of the hope.
What the events of early 2026 establish, taken together, is that the problem has finally become legible. The clinical analysis named the mechanism. The reporting documented its consequences in the lives of real people. And the jury, in a courthouse in Santa Fe, did the thing that had never been done: it looked at the design and assigned responsibility for it. None of this fixes anything by itself. The verdict will be appealed, the litigation will grind on for years, the platforms will resist every constraint on the optimisation that makes them rich. But the question the whole episode poses is no longer abstract or speculative. It is a practical question with practical answers, about objective functions and friction and valence constraints, about liability and regulation and the standing to demand better.
The deepest stake is the one that is hardest to put on a balance sheet. Empathy, the developmental psychologists insist, is not a private virtue or a personal temperament. It is the foundation of moral reasoning, of the recognition that another person's suffering makes a claim on us. A society that loses the capacity to feel that claim does not become more rational or better informed. It becomes, in Bushman and Anderson's exact and terrible phrase, comfortably numb, which is the precondition for permitting almost anything. The systems through which hundreds of millions of people now experience the world were not built to produce that outcome. But they were built without a constraint that would prevent it, and so they produce it anyway, with the same indifference that any optimisation process brings to whatever it has not been told to protect. The work ahead, legal and clinical and technical and personal all at once, is to tell it. The New Mexico jury has shown that the telling can begin in a courtroom. Whether it ends in a better machine, or merely in a long record of damages paid after the fact, is the wager now being placed.
