The Mind-Weapon Panic: From CIA Psychics and UFOs to AI Apocalypse
In September 2026, a 27-year-old researcher named Jacob Coxon resigned from Anthropic and posted a series of warnings on 𝕏. He wrote that the people building today’s most advanced AI earnestly believe it could kill us all by the end of the decade. By the time Axios reported on his resignation, his post had been viewed more than 115 million times.
Researchers had been saying similar things for years. What made Coxon’s warning unusual was that he had walked away from real money to say it. He left Anthropic two months before his equity was due to vest, forfeiting a potentially significant payoff. “I left before any of my equity vested,” he told Axios. “I no longer have anything to gain by juicing up Anthropic’s valuation.”
Coxon was not alone. Evan Hubinger, who leads Anthropic’s Alignment Science team, publicly agreed. “We really do earnestly believe AI could kill all humans,” Hubinger wrote. “I personally think it is >10 percent within the next decade.” Bilal Chughtai, a former Google DeepMind safety engineer, resigned in July 2026, saying he believed AI could kill us all. Joe Benton, another former Anthropic safety team member, raised similar alarms. They all had direct experience inside the labs.
Military secrecy generates mythology that sometimes provides useful cover for military secrecy and increased defense spending.
These discussions caught my attention as I was finalizing my two-year investigation into state-sponsored paranormal programs and UFO research in the Soviet Union. The deeper I went into the archives, the harder it became to distinguish evidence from interpretation. Genuine unexplained observations sit beside classified aerospace programs. Scientific investigations co-exist with disinformation. Military secrecy generates mythology that sometimes provides useful cover for military secrecy and increased defense spending. To muddy the waters, institutions often mix genuine information with fabricated facts, making it almost impossible for an independent observer to see reality for what it is.
The information environment has only grown more complex since the end of the Cold War, overwhelmed by constructed narratives, conspiracy theories, declining institutional trust, and competing versions of reality. The response to AI whistleblowers was at once immediate and fractured. Some members of Congress demanded new legal safeguards for AI development, while others were not so convinced.
Critics pointed to a funding network. Dustin Moskovitz, an early investor in Anthropic, is a member of the board of Coefficient Giving, the grant-making organization formerly known as Open Philanthropy. Coefficient is a major funder of the Tarbell Centre for AI Journalism, which supports reporters at large news outlets. It also funds METR, an independent evaluator that has worked with Anthropic. These interlinking connections invite criticism, even though funding alone does not prove that a journalist or evaluator has given up editorial or professional independence.
Anthropic has poured $40 million through two donations to the super PAC Public First Action, which is pushing Congress to impose new regulatory guardrails on AI. Its lobbying disclosure in July 2026 showed $2 million in just three months, $400,000 more than its April report. “You get a benefit from regulation because it hinders your competitors,” said James Andrew Lewis, a former U.S. diplomat. Of Anthropic co-founder Dario Amodei, he said: “He might actually believe some of this stuff, but I’d say self-interest really explains it more.” The lobbying fight raises a broader question: how can we tell a serious warning from a narrative that grows stronger as institutions respond to it?
The AI debate is not the first time a great power has been frightened into action by a threat it could not verify. In fact, the pattern I see resembles another feedback loop I came across in my research that started with a disputed account of an American submarine. In July 1958, the USS Nautilus became the first vessel in history to reach the North Pole using its own power, travelling submerged beneath the Arctic ice. For the Soviet Union, this was a strategic shock. American submarines could now approach unnoticed and lurk practically on its doorstep.
In December 1959, a French magazine called Constellation published an article named “Thought Transfer as a Weapon of War.” The author, Jacques Bergier, described a secret American psi program successfully tested during the Nautilus mission. A strange passenger named Jones boarded the submarine and, over sixteen days, purportedly received images telepathically from a sender at the Westinghouse laboratory in Friendship, Maryland, with a claimed 70 percent accuracy.
The story was likely fabricated. The U.S. Navy denied it. The dates didn’t work, as on the day the experiment was supposed to begin, the Nautilus was in dry dock in Portsmouth. Bergier may have been deliberately fed disinformation, or he may have invented the story himself. The U.S. was deploying new ocean sound detectors at the time and may have wanted to distract Soviet attention from that project.
In Moscow, however, no one was laughing. Professor Leonid Vasiliev, who had conducted telepathy experiments in the 1930s before Stalin shut them down, reported to his colleagues that the United States had “adopted telepathy as a means of communication with submarines.” In March 1960, cybernetics specialist Lieutenant Colonel Igor Poletaev sent an urgent report to Defense Minister Rodion Malinovsky warning of a “new, unknown psychological weapon.” The Soviet government immediately funded Vasiliev’s Laboratory for the Study of Telepathy at Leningrad State University.
Vasiliev’s 1962 book, Experiments in Mental Suggestion, became a foundational text of the new Soviet approach. Other laboratories followed. Dr. Genady Sergeyev, working out of a military-affiliated institute, developed sensitive non-contact sensors to study Nina Kulagina, a woman who claimed to move objects by psychokinesis, and filmed her deflecting compass needles while recording changes in her electrical brain activity (James “The Amazing” Randi subsequently explained how she might have performed these feats using magic). At the A.S. Popov Radio and Technical Institute in Moscow, engineer I.M. Kogan attempted to model telepathy mathematically as ultra-longwave electromagnetic radiation, treating the human nervous system as a biological radio transmitter.
The Americans, meanwhile, watched the Soviet activity with growing alarm. A 1972 Defense Intelligence Agency report titled Controlled Offensive Behavior—USSR claimed that the Soviet Union had at least twenty parapsychology centres, was spending the equivalent of $80 to $100 million annually on related research, and would soon be able to read secret documents, influence the minds of U.S. leaders, and cause “instant death” from a distance. The report triggered panic in Washington.
In response, the CIA launched Project SCANATE at the Stanford Research Institute, employing physicists Russell Targ and Harold Puthoff. They began testing subjects like the artist Ingo Swann and the stage magician Uri Geller. The program evolved through a succession of codenames (Gondola Wish, Grill Flame, Center Lane, and Sun Streak) before finally becoming Project Stargate. Its most celebrated success came in 1979, when a remote viewer named Joe McMoneagle was given coordinates for a building near a large body of water in northern Russia. He described a massive submarine under construction, with a long flat deck, a double hull, and twenty missile tubes, a vessel so large that conventional analysts insisted it could never be launched from that location. He predicted the Russians would blast a channel to the sea within months. In January 1980, spy satellite photos confirmed the launch of the largest submarine ever observed—the Typhoon class—exactly as McMoneagle had sketched.
When the KGB learned that the Americans had obtained such detailed intelligence without any physical agents in place, the conclusion was inescapable: the United States had mastered remote viewing and was forty years ahead of the USSR. The response followed in 1989, when the Soviet General Staff authorized the creation of Military Unit 10003, an expert-analytical directorate under Lieutenant General Alexey Savin. He was not the kind of man one associated with telepathy and alien contact. He had graduated from the Bauman Moscow Higher Technical School, one of the country’s most prestigious engineering institutions. He had spent his early career working on classified defense projects, held a doctorate in technical sciences, and also studied philosophy, a combination that shaped his unique approach.
Effective skepticism involves the capacity to separate the phenomenon from the information system that builds up around it.
Savin was not a cynic. He believed the human brain could access what he called the “Universal Information Field,” a planetary field of consciousness first theorized by Vladimir Vernadsky in the 1920s. He rejected Western occultism as elitist and individualistic. He called his method “metacontact,” a secularized form of military shamanism. It drew on yoga, Buddhist meditation, qigong, and Siberian trance techniques, stripped of their spiritual content and repackaged as operational methodology.
According to the general, the unit’s operational results in Extra-Sensory Perception (ESP) were truly extraordinary. Savin said his operators had remotely viewed a captured American pilot during the 1991 Gulf War. He also claimed they predicted a 1991 Kamchatka earthquake before planned military exercises, which helped the authorities to evacuate hundreds of people from the seismic zone. In another rumoured case, his operators prevented a nuclear catastrophe near Glasgow, Scotland: after remotely viewing the potential future event, they sent their predictions to London via unofficial channels. However, the most remarkable experience associated with Unit 10003 was their alleged interaction with a non-human intelligence, including an exchange of information during organized telepathic sessions. This last claim is directly connected to modern-day narratives that link psi research and UFO folklore.
In the early 1990s, the U.S.–Russian relationship experienced a period of détente. Dr. Edwin May, the last director of Stargate, first travelled to Moscow in 1992 with his colleague Larisa Vilenskaya to study ESP research. They met Savin and his team. The initial suspicion gradually melted. In May 2000, Vilenskaya and McMoneagle visited Unit 10003’s headquarters in Moscow. In one account of the visit, a Russian psychic named Klimova was asked to replicate a drawing an American made in another room; the two images were said to be mirror images.
Savin later spoke warmly of the Americans. They shared stories, exchanged methodologies, and eventually contributed to a Russian book, Psi Wars: West and East, credited to Edwin May, Joe McMoneagle, Alexey Savin, Boris Ratnikov, and Viktor Rubel. General Nikolai Sham, the former deputy chairman of the KGB, wrote the foreword. Savin even proposed a joint Russian-American centre to combat terrorism and extremism using their methods.
Edwin May’s interview on the Shawn Ryan Show, released on July 22, 2024, offered his account of the connection. May recalled that Unit 10003 had about 120 remote viewers, ten times more than Project Stargate at any given time. He said Savin’s proposal for a joint U.S.–Russian program to combat terrorism appealed to him. He wrote a detailed 30-page report and handed it to a three-star general at the Defense Intelligence Agency (DIA) in the Pentagon. May recalled the general calling the proposal “ideal” because it was cheap and put no one in harm’s way. May then joked that he could practically hear the paper going into a “burn bag.”
Project Stargate was shut down in 1995 after a CIA-commissioned assessment found that remote viewing had not shown value as a source of actionable intelligence. Military Unit 10003 was disbanded in 2003. The official reason given was that it had completed its assigned tasks. However, the decision followed years of public criticism from the Commission for Combating Pseudoscience and Falsification of Scientific Research, founded in 1998 by Nobel laureate Vitaly Ginzburg, who was quoted in a Russian newspaper, Izvestia, as saying, “Telepathy does not exist. Remote viewing is a fraud. The sooner we stop funding this nonsense, the better.”
What complicates a complete dismissal of American and Soviet psi warriors is that Savin and May appear sincere about their methods, and seem willing to set aside Cold War stereotypes about each other. Jon Ronson, the author of The Men Who Stare at Goats (subsequently made into a major motion picture), recently told the Danny Jones podcast that McMoneagle struck him as sensible, humble, and not a charlatan. Ronson also said he had no direct evidence that remote viewing worked.
Overall, the modern scientific consensus on ESP is that a few isolated experiments and case reports tend to be small, hard to replicate, vulnerable to methodological bias, and they often shrink or disappear under stricter controls. Ultimately, both Project Stargate and Unit 10003 developed in response to concerns about each other’s perceived progress. Although some of the claims they acted on were exaggerated or poorly tested, the programs they helped build were genuine and at some point taken seriously by their governments.
The current AI debate seems to be centered on similar alarmist views. Concern about China’s progress can shape American investment and policy even when its implications remain uncertain. Coxon himself told Axios that fears about both OpenAI and China can become excessive and help justify moving faster. That does not settle the question of AI’s dangers. It shows how competition can turn uncertainty into pressure to act.
The real challenge is separating genuine evidence of danger from the pressure generated as competitors, investors, laboratories, and advocates react to each other’s fears.
The same scrutiny should apply to policy and funding. In June 2026, Dario Amodei, CEO of Anthropic, published a policy article in favor of mandatory third-party safety tests for frontier AI models. Such tests would address real dangers; however, they would also benefit well-resourced and well-established companies. Licensing conditions and minimum compute requirements could have a similar effect. A single rule can serve as both a safety measure and a competitive advantage. Therefore, it is important to find out who sets the standards, who can meet them, and how independent assessors are selected. Similar questions arise around the funding chain. Dustin Moskovitz’s involvement, Coefficient Giving’s support of the Tarbell Centre, and the funding connections with AI evaluation all involve potential conflicts that should be made public and investigated. Although Tarbell claims that its funders have no editorial control, a network is not a conspiracy, and it still plays a role in determining which warnings receive funding and reach a wider audience.
At the same time, one doesn’t have to be excessively cynical about this, or jump to unwarranted conclusions. Coxon had given up his unvested equity, and his warning ought to be taken seriously. Max Nadeau from Coefficient Giving echoed Coxon’s message shortly after it came out. That fact does not prove coordination; it only shows how quickly a well-connected network can spread a warning that addresses its own concerns. Sincerity and strategic value can co-exist. This was also the case of the psi programs. Savin and May seem sincere and have established real institutions. Their sincerity did not prevent a self-reinforcing threat narrative from emerging. On the contrary, it helped make the feedback loop believable.
One important difference is that although remote viewing remains of doubtful practical value, AI systems have already begun to carry out useful tasks, and their capabilities continue to develop. The consequences are more serious, and the pressure to act quickly is greater. Coxon’s warning is not that artificial intelligence will suddenly come to life and attack us; instead, businesses are accelerating the development of systems that are not only more powerful but also harder to understand or control. That is a legitimate concern. The real challenge is separating genuine evidence of danger from the pressure generated as competitors, investors, laboratories, and advocates react to each other’s fears.
Like the AI panic and psi wars, the modern-day Unidentified Anomalous Phenomena (UAP) disclosure movement in the U.S., which began in 2017 with The New York Times article disclosing the Pentagon’s Advanced Aerospace Threat Identification Program (AATIP), and continued through whistleblowers and U.S. congressional hearings, is yet another example of an unstable information environment that mixes real institutional process with sensational claims, ideological and religious belief systems, state secrecy, funding, and rival narratives.
In the end, it is not simply a question of deciding in advance that psi is impossible, that UAPs are extraordinary, or that artificial intelligence poses existential risks. On the contrary, effective skepticism involves the capacity to separate the phenomenon from the information system that builds up around it. Even when sincerity is strongly felt, it does not count as evidence; people taking part in these feedback loops may be completely convinced of their own stories, and this conviction can give the narrative an appearance of plausibility. The real challenge is to suspend immediate judgment and examine what the phenomenon can reveal once the surrounding narrative is stripped away.