To prepare for your return, try Grand Continent . The ideas and analyses you won’t find anywhere else. But you’ll soon see them everywhere. In your inbox from 8 euros a month
As the Yale statistician Edward Tufte once noted: “There are only two industries that label their customers ‘users’: drug dealers and software manufacturers.” On August 27, 2026, Meta signed an amicable agreement with 47 American states that were suing it for having “deliberately endangered the youngest” with its social media platforms. To understand precisely what Meta was accused of and what lies at the heart of its algorithmic machinery, we must take a detour that has not often been offered since, not by law, but by neuroscience.
What is Meta accused of?
First, a double discourse. Meta knew its products were dangerous thanks to internal studies, yet it never publicly acknowledged this risk, instead assuring users that they faced no danger. One can thus liken this to cigarette makers who, through their internal studies, knew their products were harmful but long claimed they were beneficial to health, or to Takata airbags, which also knew their products were dangerous but remained silent and continued selling them without informing of the hazards.
Meta did not act out of malice or a love of harm. As whistleblower Frances Haugen, the originator of the Facebook Papers, has noted: “What was good for its business was bad for its users, and Mark Zuckerberg systematically weighed in favor of the business while knowing it would come at the expense of users’ mental health.”
From existential fear to cognitive predation
We know that Mark Zuckerberg was terrified by a nightmare: MSN Messenger and MySpace. He did not deliberately harm users’ mental health because he enjoys chaos, but because he was absolutely dread-stricken by the idea that people, especially the youngest, would gradually abandon his app to spend more time on rival platforms. He lived in a constant fear of suffering the same fate as MySpace or MSN Messenger, the early social networks that were at one time the ultimate venues for digital social life among the young, before being gradually replaced by Facebook.
So Zuckerberg acted on two fronts. On the one hand, he exhibited what we might call “passive predation” by opposing any safety measures strongly suggested by his own research teams. These measures, designed to create friction to reduce time spent and “engagement,” were categorically rejected by Zuckerberg for fear that users would simply depart for the competition, accelerating a decline similar to MySpace’s.
He also demonstrated “active predation” by turning to neuroscience to optimize the dopaminergic response triggered by using his platforms in the brains of his younger users. In plain terms, Mark Zuckerberg deliberately used reinforcement learning to flood their brains with dopamine, drastically reducing their ability to resist the urge to return again and again to his platforms.
Using neuroscience to deliberately exploit the vulnerabilities of young brains
We must remind ourselves that the term “addiction” in the strict sense used to describe a relationship to social networks and screens — in a sense equivalent to that used for chemical drugs or sports betting — is still a matter of debate within the medical community and is not universally accepted today.
Yet neuroscience reveals a decisive element.
Mark Zuckerberg and his research teams knew that adolescents show different rates of development across brain regions. In simple terms, the reward-processing region of the brain is the nucleus accumbens, while the region that evaluates long-term risks and exercises impulse control is the prefrontal cortex.
It turns out that in young people the nucleus accumbens develops much more quickly than the prefrontal cortex, creating a heightened susceptibility to rewards and a greater difficulty in regulating impulsive behaviors and sudden cravings. Zuckerberg thus “logically” relied on a predatory design that optimized reward patterns to exploit this weakness.
Our understanding of how reward mechanisms shape the human brain rests in particular on the behavioral sciences and neuroscience work of B. F. Skinner, a Harvard psychology professor who published in 1957 with Charles Ferster the book Schedules of Reinforcement, and on Wolfram Schultz, a Cambridge neuroscientist who published in 1998 a key article entitled “Dopamine neurons and reward prediction error.”
Together they show that, contrary to many common beliefs, dopamine is not a neurotransmitter signaling pleasure or reward by itself. Dopaminergic neurons, those that secrete dopamine, are activated or blocked by the prediction of whether a reward will arrive and by whether that prediction comes true. It is an anti-intuitive but absolutely essential point for understanding Meta’s strategy. Reward prediction matters as much to the brain as the reward itself.
One sometimes hears that “a euro won on a sports bet is worth more than three euros earned by working.” We now know there is a clear neuroscientific basis for this real phenomenon. If the brain predicts a certain reward and then receives it, dopaminergic neurons eventually stop reacting. We can extrapolate this result and explain many human traits: the brain ceases to fully enjoy a reward once its arrival was certain. The brain is much less responsive to what it has “taken for granted.” Dopaminergic neurons fire most when anticipating a reward and during uncertainty.
The dopamine reward circuits are thus closely linked to the concept of reinforcement learning, the very mechanism used to train many AI systems that underlie “neural networks.” Dopamine and the reward magnitude are maximal when the brain knows a reward can come at any moment but lacks a clear model to predict when. If a reward occurs, for example, at a predictable interval (every hour), after a few hours the dopaminergic neurons stop reacting. In contrast, if rewards come irregularly — sometimes after ten minutes, sometimes after two hours, sometimes after forty minutes — the dopaminergic neurons remain hyperactive with each reward.
Structure of the active predatory design
Fully aware of these scientific discoveries, Mark Zuckerberg deliberately introduced into his algorithms a “predation philosophy” aimed at maximizing the impact of every “reward” on the dopaminergic neurons, and thus on the nucleus accumbens, the brain region that receives them.
Since in young people the nucleus accumbens is more developed than the prefrontal cortex, which allows them to control impulses and “temper things,” these dopamine surges can be understood as intermittent reinforcement, a form of learning. Zuckerberg thus integrated into his algorithms a predation philosophy that plays with the brains of young people in the same way researchers play with the brains of mice. It is reasonable to call this “active predation” because it is not simply, as we will see below, a matter of opposing safety measures, but of steering the design to deliberately target the still-developing brains of the most young and vulnerable users.
Concretely, what did he do? Three key aspects stand out: the timing of notifications, the infinite scroll and its “slot machine” effect, and the aim of provoking “school blasts” to maximize usage during class times.
Meta’s algorithms deliberately delay notifications to make their arrival harder to predict, thereby maximizing their brain impact. Imagine a teenager named Ellie posting a photo on Instagram. Ellie would receive likes on that photo at a regular pace of about one every five minutes. Instagram’s algorithm would not send a notification every 5, 10, or 15 minutes because that regular rhythm would make the arrival of the reward predictable and thus less effective on the dopaminergic neurons. The algorithm would instead send a notification after 5 minutes, then after an hour, then after 27 minutes, and so on.
The brain then understands that a reward can arrive at any moment, with no precise forecast. It is precisely this uncertainty that keeps the reward circuit alert and reinforces the response: a massive secretion of dopamine. The irregularity of notifications reproduces the principle of intermittent reinforcement, whereby an unpredictable reward motivates repetition of a behavior. But unpredictability does not mean randomness. The moment of the notification can be calculated to maximize the chances that the user returns to the platform.
The algorithm analyzes the user’s entire usage data to optimize when notifications are delivered. Returning to Ellie’s case, the algorithm measures the average duration of her sessions and predicts the moment she is most likely to leave. If she gets many likes on her photo, the app will not indicate it to her and will wait precisely for the moment she is most likely to depart before sending the notification and thus the “reward.”
A second aspect of this “active predation philosophy” concerns the infinite scroll and, more specifically, the gesture of pulling the feed down to refresh. Meta researchers themselves compared this mechanism to a slot machine: the user acts without knowing what will appear, but anticipates the possibility of a dopamine trigger.
Meta did not invent the infinite scroll, conceived by Aza Raskin in 2006, nor the pull-to-refresh mechanism, developed by Loren Brichter in 2009. Mark Zuckerberg’s company, however, adopted and optimized these technologies to strengthen their effect on the brain’s reward circuit.
On one hand, they added a circular arrow that spins during the feed refresh even when the network is excellent and there is no loading time, imitating the rotation of a slot machine that is essential to mobilize the dopaminergic neurons. Indeed, dopamine secretion peaks when that wheel spins and the brain anticipates a potential reward but without certainty of its arrival. This rotating wheel during content refresh is thus added and extended artificially, regardless of network state, to mobilize dopaminergic neurons for the sole purpose of triggering reward-seeking urges.
Within this “slot machine effect,” the algorithm also delivers “rewards” intermittently and unpredictably: sometimes you’ll get a mediocre feed filled with dull advertisements—“no reward”—and at other times a feed with a viral video aligned with your interests—a “small reward”—and other times a feed with a fake news item you’ll be happy to believe because it confirms your beliefs and thus constitutes a “great reward.”
Finally, we can talk about Meta’s aggressive pursuit of deeper use among teenagers during school hours. In 2017, Mark Zuckerberg made his stance crystal clear: “Our key objective for 2017 is for teenagers to spend as much time as possible on our platforms.” It is thus that Meta’s teams began designing what they called “school blasts”: waves of notifications sent to youths whose location data indicated they were inside a school building, aimed at leveraging peer pressure and the fear of missing out to spark widespread use during class. A Meta executive even wrote in an internal email: “We must design so that every student cannot skip a chemistry class without glancing at their phone.”
Passive predatory design: nothing must stand in the way of online time
We can then examine what might be termed “passive predation,” the deliberate refusal to implement safety improvements for users. One can summon the Ford Pinto example: a car that Ford knew was dangerous. It would have cost only $11 per car to save lives, but an internal study concluded that doing nothing would be cheaper for the company. In the same vein, Mark Zuckerberg did nothing and allowed the mental health of many adolescents to deteriorate, sometimes quite severely.
This “passive predation” should not be viewed as any less serious, because it constitutes a decisive refusal to address user safety and well-being. Almost all safety measures suggested to Mark Zuckerberg by his researchers were adopted only after internal studies demonstrated the very serious harm Meta’s platforms could cause young users.
The first aspect of passive predation concerns the manipulation of minors by adults. As early as 2019, Meta knew that adults were contacting adolescents and children they did not know on Instagram and was urged to act. But its own Ford Pinto-like internal calculus — evaluating the “cost to the company of increasing user safety” — estimated that making teenage accounts private by default, thus preventing unknown adults from talking to them, would remove 5.4 million unwanted interactions per day, which would be very bad for growth and engagement. It was therefore only implemented in 2024, leaving billions of interactions between adults and children or teenagers to unfold in the meantime.
The second aspect of Meta’s passive predation concerns the spread of fake news, misinformation, and hate speech. Meta’s own research teams suggested various mechanisms to curb this phenomenon, grouped under the term “friction the virality.”
The idea was to slow the viral spread of content because it contains an overrepresentation of fake news and hateful material. For example, by limiting how many times content can be shared or by requiring the user to comment before sharing again. Even when an internal study showed that these mechanisms could significantly reduce the proportion of fake news, Mark Zuckerberg personally opposed any friction that would reduce engagement and risk driving users to rival platforms.
It’s worth noting that Meta’s teams interviewed European policymakers who admitted feeling compelled to post toxic, even hateful content to gain at least some visibility. Some explicitly asked Meta to modify its algorithms, notably to reduce the promotion of content that provoked anger so as not to have to produce a hateful discourse.
The third aspect of Meta’s passive predation concerns time spent on the app and the deliberate refusal to introduce any friction. Psychological research has highlighted the benefits of “stop cues” in media consumption. New York University psychologist Adam Alter, who has extensively studied “addictive” technology and published in 2017 the book Irresistible: The Rise of Addictive Technology and the Business of Keeping Us Hooked, underlined the importance of these stop cues in the fight against addiction. He distinguishes between 20th-century media that contained such cues (end of a chapter in a book, bottom of a magazine page, end of a TV program, or a commercial) and today’s tech UX, which deliberately omits them by normalizing infinite scrolling. Meta’s internal studies repeatedly established a clear link between this system and compulsive use, sleep disturbances, and body image issues among adolescents.
How do we know all this?
All of this is known thanks to whistleblower Frances Haugen, who left Meta taking tens of thousands of compromising documents with her, published notably through the Facebook Papers in 2021, and thanks to parliamentary inquiries launched by Congress. Lawmakers heard from many current and former Meta officials. Then, the lawsuits filed by American states against Meta triggered investigations and yielded numerous search warrants and orders to produce documents.
Where does UX end and predation begin?
What Meta is accused of is the design and user experience of its social networks, especially Facebook and Instagram. But is there not, somewhere, a paradox in blaming a company whose core business is precisely to create the most pleasant social networks possible for being “too effective” at it? Do we blame a car manufacturer for making cars “too pleasant to drive”? After all, a term and a whole field of analysis were created to describe the optimization of design to make the experience as enjoyable as possible: UX, or User Experience. So what is the difference between an excellent UX that deserves applause and a design that falls under “predation”?
One could compare a social platform with an “excellent user experience” and a design that makes you want to spend time there to a cyclist who wins the Tour de France. Meta, whose design is described as “predatory,” would resemble a cyclist who is fully doped and wins the Tour de France only through repeated abuses endangering his own health. In one case, the competition is won “legitimately” by targeting fully trained adults with quality content. In the other, one wins by cheating and using an unfair advantage: targeting young, particularly vulnerable minds by exploiting neuroscience.
Do we have to choose between safety and competitiveness?
The evidence shows a fundamental absence of social ethics and epistemic responsibility in Mark Zuckerberg, his deliberate refusal to accept any responsibility for users’ health, and his willingness to harm them as long as it benefits business.
Mark Zuckerberg opposed almost all safety measures suggested by his research teams in the name of his company’s competitiveness. As noted earlier, his reasoning was simple: “If we introduce friction on our platforms, our users will choose the path of least resistance and simply go to our competitors.” This is the same argument farmers use to oppose pesticide bans and to request their reintroduction: “If we don’t do it, others will, they’ll be more competitive than us, and the public will inevitably suffer the harmful health effects, but only we will bear the economic consequences.”
These are precisely the same arguments currently used by the tech giants to resist strict AI regulation: “If we do this, China will not do it and will leap ahead.” These are arguments we can readily counter if we place public health at the heart of our strategies and political thinking.