There was a time when the suggestion that an intelligence agency might secretly experiment on human beings in an attempt to understand, disrupt or manipulate the mind sounded like the raw material of paranoid fiction, yet during the Cold War the United States Central Intelligence Agency created a research program whose surviving records would eventually reveal experiments involving psychoactive drugs, hypnosis, behavioral modification and unsuspecting human subjects, leaving behind one of the most disturbing questions in modern history: how far will institutions go when they become convinced that understanding the human mind is a matter of national security?
More than seventy years later, the laboratories have changed almost beyond recognition, because the glass of LSD, the hidden safe house and the psychiatrist observing a confused patient through a one-way mirror belong to another technological age, while billions of people now spend hours every day inside digital environments where invisible systems decide which stories rise to the top of a screen, which disappear beneath thousands of others, which emotions receive the greatest amplification and which message appears at precisely the moment when a particular person may be most receptive to it.
There is no credible evidence that artificial intelligence, social-media algorithms or today’s technology companies represent a secret continuation of MKUltra, and treating that claim as established fact would reduce a fascinating historical comparison to another conspiracy theory, but the absence of such a connection does not make the comparison meaningless, because MKUltra and the algorithmic age are separated by completely different technologies while remaining connected by a question that has never entirely disappeared: if human perception can be altered by changing what reaches the mind, how much influence can be exercised before the person being influenced realizes that anything unusual is happening at all?
THE COLD WAR SEARCH FOR A DOORWAY INTO THE HUMAN MIND
Project MKUltra emerged during an atmosphere of intense Cold War anxiety in which American intelligence officials feared that communist governments might have discovered sophisticated methods of interrogation, psychological conditioning or so-called brainwashing, and the possibility that Soviet or Chinese researchers possessed techniques capable of breaking resistance encouraged U.S. agencies to investigate drugs, hypnosis and other methods that might reveal whether human behavior could be deliberately modified.
What began in 1953 was not one neatly organized experiment but a sprawling collection of projects that reached into universities, hospitals, research foundations and other institutions, often through funding arrangements designed to obscure the CIA’s involvement, while many researchers involved in individual studies apparently did not know who ultimately stood behind the money. When previously overlooked financial records were discovered during the 1970s, the U.S. Senate investigation identified 149 MKUltra subprojects, along with 185 nongovernmental researchers and assistants and 80 institutions associated with the recovered material, while the documents themselves referred to research involving behavioral modification, drug acquisition, human testing and the surreptitious administration of drugs.
The numbers alone make the program remarkable, but they do not explain why MKUltra became so deeply embedded in modern conspiracy culture, because what transformed it from an obscure intelligence project into something almost legendary was the discovery that at least some people had become experimental subjects without knowing they were participating in an experiment at all.
During the 1977 Senate hearings, CIA Director Stansfield Turner acknowledged evidence supporting unwitting testing in several of the recovered MKUltra subprojects, while testimony concerning CIA safe houses described individuals encountered in bars and subsequently brought to locations where drugs could be tested under covert observation; the surviving records were incomplete, and witnesses repeatedly warned senators not to infer more than the documentation could prove, yet the central fact was no longer disputable, because the CIA had conducted experiments in which unsuspecting people were exposed to drugs while their reactions were observed.
WHEN LSD BECAME AN INTELLIGENCE TOOL
Among the many substances explored during this period, LSD occupied a special place in the imagination of intelligence researchers because its ability to profoundly alter perception, emotion and a person’s sense of reality suggested possibilities that seemed extraordinary in an age obsessed with interrogation and psychological warfare, particularly if a chemical could produce confusion, weaken resistance or temporarily destabilize someone’s confidence in his own senses.
The disturbing element was not simply that LSD was studied, because the substance was also the subject of legitimate psychiatric and scientific interest during the same era, but that the CIA wanted to understand what happened when a person received such a drug without knowing that it had been administered, a requirement that transformed some experiments from questionable research into profound violations of informed consent.
The notorious CIA safe-house operations later associated with the name Operation Midnight Climax grew from this logic, because the agency wanted environments in which unwitting subjects could be observed while remaining unaware that anything resembling a government experiment was taking place, and testimony before the Senate described safe houses equipped for covert observation, including references to two-way mirrors and individuals brought in after contacts made in social settings. Although decades of retelling have added sensational claims that are not always supported by surviving documents, the official hearing record itself confirms enough to make embellishment unnecessary.
The real horror of MKUltra therefore lies less in the mythology of a perfect mind-control drug, which the program never produced, than in a much more believable and historically documented process through which researchers gradually accepted the idea that the mind of another human being could become an experimental territory without that person’s knowledge.
FRANK OLSON AND THE EXPERIMENT THAT ENDED IN A HOTEL WINDOW
No episode illustrates the danger of that philosophy more powerfully than the death of Frank Olson, a U.S. Army scientist who worked in biological research and attended a gathering of Army and CIA personnel in November 1953, where LSD was secretly placed in a drink consumed by Olson and several other men as part of an experiment whose full purpose they had not been told in advance.
According to the congressional record, colleagues described Olson before the experiment as professionally respected and apparently psychologically normal, yet after receiving LSD he became increasingly distressed, agitated and troubled, eventually requiring medical attention and traveling to New York with CIA-associated personnel for consultations with Harold Abramson, a physician experienced with LSD whose research had received indirect CIA support.
During the early hours of November 28, after days of psychological turmoil, Olson went through the closed window of a tenth-floor room at New York’s Statler Hotel and fell to his death, while the CIA subsequently concealed important circumstances surrounding the experiment from his family and attempted to prevent the death from becoming connected publicly with the agency. The Senate investigation later reconstructed much of this sequence from internal memoranda and testimony, although arguments over whether Olson died by suicide or under more sinister circumstances continued for decades and produced theories that extend beyond what the surviving evidence can firmly establish.
The case remains chilling precisely because one does not need to accept the most dramatic later theories to understand its significance, since the documented sequence is extraordinary enough: a scientist was unknowingly given a powerful hallucinogen during an intelligence experiment, his psychological condition deteriorated severely, and eight days later he was dead.
DONALD EWEN CAMERON AND THE ATTEMPT TO BREAK PERSONALITY BEFORE REBUILDING IT
If Frank Olson’s story reveals what could happen when an unsuspecting person encountered a psychoactive drug, the experiments associated with psychiatrist Donald Ewen Cameron at Montreal’s Allan Memorial Institute reveal an even more ambitious idea, because Cameron became interested in treatments intended to disrupt existing patterns of thought before attempting to construct new psychological responses.
McGill University’s archival record describes the Montreal experiments conducted at the Allan Memorial Institute between 1957 and 1964 and identifies methods including “psychic driving,” drug-induced sleep, electroconvulsive therapy and sensory deprivation, while also confirming that the work received CIA funding as part of MKUltra.
Cameron’s concept of psychic driving involved repeated recorded messages intended to influence patients after existing psychological patterns had been weakened through other intensive treatments, and although the work was presented within a psychiatric framework rather than as a cinematic attempt to create obedient agents, its connection with CIA behavioral research transformed it into one of the darkest chapters associated with MKUltra, particularly because many of the people subjected to these procedures were already vulnerable psychiatric patients who could not have understood the intelligence context surrounding the funding.
The significance of Cameron’s work extends beyond the specific techniques used, because it demonstrates how the search for influence could move from drugs toward something conceptually broader: the attempt to alter the informational and psychological environment surrounding a human being until ordinary patterns of perception and response began to change.
THE FILES THAT VANISHED AND THE MYSTERY THEY LEFT BEHIND
Much of what the public now knows about MKUltra survived almost by accident, because extensive program records were destroyed in 1973, while the financial files later uncovered had escaped destruction largely because they had been stored separately from material researchers initially searched for during earlier investigations. The Senate hearing noted that the newly discovered folders were primarily financial records containing vouchers, approvals and accounting documents rather than complete operational reports, which meant investigators could reconstruct the scope of many projects without recovering every detail of what had occurred.
This destruction created the perfect environment for mythmaking, because whenever documents disappear from a secret intelligence program, the missing pages can become more powerful in the public imagination than the pages that survived, allowing almost any later theory to be projected into the gaps.
Yet those gaps should not tempt us to confuse possibility with evidence, because the strongest way to understand MKUltra is not to enlarge it into a universal explanation for every alleged mind-control story but to recognize that the documented history is already unsettling enough: institutions with enormous resources genuinely studied behavioral modification, secretly administered drugs to some people, concealed sponsorship of research and later destroyed significant portions of the record.
WHEN THE LABORATORY MOVED FROM THE HOSPITAL TO THE SCREEN
The modern world appears completely different until we examine what digital systems actually do, because social-media platforms and recommendation engines continuously alter the informational environment surrounding their users, selecting some material for visibility while allowing other material to disappear almost entirely, then measuring the behavioral consequences through clicks, pauses, shares, comments, watch time and countless other signals.
Unlike MKUltra, this process generally has no secret intelligence officer slipping chemicals into a drink and no researcher attempting to erase personality, yet it possesses one capability Cold War experimenters could scarcely have imagined: the possibility of observing the reactions of millions of people simultaneously while changing what different groups see.
A user might assume that scrolling is a passive activity, but every decision provides information from which a system can learn, because the video watched twice may reveal more than the video liked once, the article abandoned after five seconds says something different from the article read to the end, and the content that provokes anger, fascination or fear can become statistically distinguishable from material that produces indifference.
This does not mean that an algorithm understands the human mind as a psychologist would, because it may have no concept of fear, ambition, loneliness or political identity in any human sense, yet it can still become increasingly effective at predicting which stimulus is likely to produce which response.
At that point, the difference between predicting behavior and shaping the environment in which behavior occurs becomes surprisingly difficult to define.
THE FACEBOOK EXPERIMENT THAT ALTERED 689,003 NEWS FEEDS
One of the clearest demonstrations of this new experimental scale appeared in 2014, when researchers published the results of a Facebook study involving 689,003 users, whose News Feeds had been altered during one week in 2012 so that some participants encountered less positive emotional content while others encountered less negative emotional content.
The researchers then measured the emotional language those users subsequently produced in their own posts and found small but statistically detectable changes corresponding to the emotional material they had been exposed to, providing experimental evidence that adjusting the emotional environment of a social-media feed could alter subsequent emotional expression even without direct face-to-face interaction.
The effect itself was modest, and the experiment certainly did not demonstrate that Facebook could remotely control people’s emotions, but the ethical controversy surrounding the study revealed something historically interesting, because most participants had no ordinary understanding that their information environment had been altered for a psychological experiment.
The comparison with MKUltra must remain proportionate, since reducing positive posts in a News Feed is obviously not equivalent to secretly administering LSD, yet the underlying question feels strangely familiar: what does informed consent mean when the subject does not know that the environment itself has been experimentally modified?
CAMBRIDGE ANALYTICA AND THE VALUE OF KNOWING WHAT A PERSON IS LIKE
The Cambridge Analytica scandal added another element to the emerging picture by demonstrating how apparently trivial digital behavior could be transformed into psychological profiles intended for political targeting, because according to the U.S. Federal Trade Commission, data collected through Aleksandr Kogan’s Facebook application was used to generate personality scores that were subsequently matched with American voter records for profiling and targeted advertising.
The FTC alleged that the application collected information from hundreds of thousands of direct users and tens of millions of their Facebook friends, many of whom never interacted with the application themselves, while Cambridge Analytica sought to use Facebook information to predict personality traits and provide voter profiling and microtargeting services.
Claims about Cambridge Analytica’s actual power were sometimes exaggerated after the scandal became famous, and there is no convincing evidence that its psychological profiling gave it a magical ability to determine election outcomes, but the larger lesson was difficult to dismiss: information that appears meaningless when considered individually can become extremely valuable when combined across millions of observations.
A political consultant no longer needed to ask a voter directly what frightened, motivated or angered that person if enough behavioral data could be analyzed to make a useful prediction.
THE ALGORITHM DOES NOT NEED TO TELL YOU WHAT TO BELIEVE
The most powerful form of algorithmic influence may therefore have little resemblance to direct persuasion, because a system does not necessarily need to convince someone that the world is dangerous if it can repeatedly show that person dangerous events until danger begins to feel omnipresent.
If stories about violent crime hold a user’s attention longer than stories about declining crime statistics, the algorithm may gradually provide more robberies, assaults and frightening footage without ever making the explicit statement that society is collapsing, while the person exposed to that stream of information may slowly develop a perception of the world based not on everything that exists but on everything the system has learned will keep that particular person watching.
The same mechanism can operate around politics, immigration, war, health, conspiracy theories or social conflict, because algorithms optimized for engagement do not necessarily require ideological intentions when emotionally charged material happens to perform better than calm explanation.
This creates an unsettling possibility in which nobody needs to sit inside a secret control room instructing the machine to amplify fear, anger or division, because those characteristics can emerge simply because they generate measurable reactions.
THE 2026 X EXPERIMENT: WHAT HAPPENS WHEN THE FEED CHANGES?
In February 2026, researchers publishing in Nature reported a seven-week field experiment involving active U.S. users of X who were randomly assigned to experience either an algorithmic feed or a chronological one, providing unusually direct evidence about what can happen when the mechanism selecting information is altered.
Among participants who moved from a chronological feed to X’s algorithmic feed, the researchers observed increased engagement as well as measurable shifts toward more conservative positions on certain policy priorities, perceptions of criminal investigations involving Donald Trump and views concerning the war in Ukraine, while users exposed to the algorithmic feed also became more likely to follow conservative political activist accounts that could continue shaping their information environment afterward. The authors emphasized important limitations, including the fact that the findings concerned one platform during one specific period and did not demonstrate comparable effects on partisanship or every political attitude, but their broader conclusion was difficult to ignore: under some real-world conditions, changing the algorithm that chooses what people see can causally influence both online behavior and political opinion.
A separate Nature study published later in 2026, involving approximately 2,000 participants using experimentally controlled feeds around the 2024 U.S. presidential election, found that engagement-based ranking amplified intergroup, moralized and emotional information and affected perceptions surrounding partisan hostility, reinforcing the idea that algorithms can alter not only which opinions become visible but also the emotional atmosphere through which people interpret the behavior of others.
This distinction matters enormously because influence does not always require changing what someone believes directly; sometimes changing what that person believes everyone else believes can reshape the social reality in which future decisions are made.
ARTIFICIAL INTELLIGENCE ADDS SOMETHING THE OLD ALGORITHMS COULD NOT DO
Recommendation systems primarily decide which existing messages people encounter, while generative artificial intelligence introduces another possibility because an advanced language model can create the persuasive message itself, adapting arguments, vocabulary and tone with a speed that human persuaders could never reproduce at comparable scale.
A 2025 study published in Nature Communications, based on three preregistered experiments involving 4,829 participants, found that people exposed to persuasive messages generated by large language models showed measurable shifts in their attitudes toward several public-policy questions compared with participants who received neutral messages, while the AI-generated persuasive material performed broadly similarly to messages produced by ordinary human writers. The effects were not enormous and did not demonstrate irresistible persuasion, yet the researchers emphasized an implication that becomes much more significant at scale: persuasive political messages can now be generated quickly, cheaply and in enormous quantities.
The twentieth-century intelligence researcher dreamed of discovering a technique powerful enough to work on almost anyone, but artificial intelligence points toward a completely different model in which there may never need to be one universal persuasive formula, because a system could theoretically generate different messages for different psychological profiles and continuously adjust them according to how audiences respond.
The old question was whether one could find the key to the human mind.
The new question is whether millions of different keys can be manufactured automatically.
MKULTRA AND AI ARE NOT THE SAME, WHICH IS EXACTLY WHY THE COMPARISON MATTERS
It would be historically careless to describe social media or artificial intelligence as “MKUltra 2.0,” because MKUltra involved intelligence agencies, covert experimentation, psychoactive drugs and profound violations of consent that belong to a very specific historical context, while contemporary algorithms generally exist within commercial platforms whose primary purposes involve recommendation, advertising, communication and engagement.
What makes the comparison meaningful is therefore not a secret institutional connection but a transformation in the philosophy of influence.
MKUltra researchers explored whether the mind could be disrupted from the outside through chemical and psychological intervention, while modern digital systems increasingly learn by observing the mind indirectly through behavior, accumulating patterns until they can predict which information is most likely to attract, frighten, persuade or retain a particular person.
One approach attempted to penetrate the mind by force.
The other can surround the mind with a customized information environment.
One searched for the perfect chemical.
The other searches continuously for the next piece of content most likely to produce a measurable response.
THE MOST DISTURBING POSSIBILITY MAY BE THAT THERE IS NO MASTER CONTROLLER
Conspiracy theories often imagine control as something centralized, intentional and secret, with a small group sitting behind a curtain and directing the thoughts of millions, yet algorithmic systems introduce a stranger possibility because harmful patterns can emerge even when nobody has deliberately designed the final outcome.
If a platform rewards engagement and the system discovers through billions of interactions that anger keeps users present longer than calm discussion, sensationalism spreads faster than careful uncertainty and frightening claims generate more comments than reassuring explanations, the algorithm can amplify precisely those characteristics without possessing a political ideology, a psychological theory or any conscious desire to manipulate society.
This is why asking only “Who controls the algorithm?” may miss the more important question, because the answer may sometimes be less revealing than asking “What is the algorithm being rewarded for?”
A system optimized for attention does not need to hate, deceive or believe anything in order to learn that particular emotional reactions are profitable.
FROM THE GLASS OF LSD TO THE GLOWING SCREEN
Frank Olson consumed a drink in 1953 without knowing that a chemical experiment was about to alter his experience of reality, while patients in Montreal entered psychiatric treatment without understanding the full intelligence context surrounding research that would later become connected publicly with MKUltra, and the surviving documents eventually revealed a world in which the pursuit of behavioral knowledge had crossed boundaries that now seem almost impossible to defend.
The twenty-first century presents a profoundly different world in which most people voluntarily carry the experimental environment with them, willingly return to it dozens of times every day and continuously provide the behavioral signals from which increasingly sophisticated systems learn, yet voluntary participation does not automatically answer every ethical question when the mechanisms guiding attention remain largely invisible to those being guided.
There is no LSD hidden inside the smartphone, no evidence that today’s AI companies secretly inherited MKUltra and no scientific basis for claiming that algorithms can transform human beings into programmable machines, but there is growing evidence that changing information feeds can influence emotional expression, alter perceptions, affect online behavior and, under certain conditions, shift political attitudes.
Perhaps the Cold War researchers were therefore searching for something more dramatic than they needed when they imagined that control over the human mind would require a miraculous chemical or an irresistible psychological technique, because the more subtle form of influence may not involve taking control of the mind itself at all.
It may be enough to control the sequence of reality placed in front of it.
If one person repeatedly sees fear, another repeatedly sees outrage and a third repeatedly sees confirmation of everything he already believes, each may eventually inhabit a slightly different world while remaining convinced that the screen has merely shown reality as it exists.
The laboratory has become invisible, the experiment may resemble ordinary life, and the person inside it does not necessarily feel controlled, manipulated or even observed.
He simply opens the phone, begins to scroll, and waits to see what the algorithm has decided should exist next.
SOURCES AND FURTHER READING
The historical material concerning Project MKUltra, its recovered 149 subprojects, researchers, institutions, unwitting experiments, CIA safe houses, Frank Olson and the destruction of records is based primarily on the 1977 Joint Hearing of the U.S. Senate Select Committee on Intelligence and the Subcommittee on Health and Scientific Research.
Information concerning Donald Ewen Cameron and the Montreal experiments is supported by the McGill University Archival Collections, which document the Allan Memorial Institute experiments and their relationship with CIA-funded MKUltra research.
The Facebook emotional-contagion experiment is documented in the peer-reviewed 2014 study “Experimental Evidence of Massive-Scale Emotional Contagion Through Social Networks,” involving 689,003 users.
Information concerning Cambridge Analytica’s Facebook data collection, personality profiling and political targeting is based on findings published by the U.S. Federal Trade Commission.
The modern discussion of algorithmic influence draws on the 2026 Nature studies examining X’s feed algorithm and engagement-based social-media ranking.
The discussion of AI-generated persuasion is based on the 2025 Nature Communications study “LLM-generated messages can persuade humans on policy issues,” involving 4,829 participants across three preregistered experiments.
