Could artificial intelligence eventually become powerful enough to escape human control, or is America being swept up in another technological panic? The warnings are no longer coming only from science-fiction writers and AI skeptics. Scientists, researchers and executives inside the companies building the world’s most advanced AI systems have publicly warned about the possibility of uncontrollable superintelligence, catastrophic misuse and even human extinction. At the same time, critics argue that AI fear can benefit Big Tech by attracting billions in investment, encouraging regulations that protect established companies and keeping the public focused on an increasingly dramatic technological race. The uncomfortable possibility is that both things could be true: the fear can be exploited, and the danger can still be real.

AI Experts Warning About Human Extinction and Loss of Control
Dr. Roman Yampolskiy, a computer science professor at the University of Louisville, coined the term “AI safety” around 2010. He spent years trying to make advanced AI controllable. He now says that once the machine is smarter than the people holding the leash, control is not difficult. It is impossible. He has put the long-term chance that superintelligence wipes out humanity as high as 99.9 percent. “If we create general superintelligences, I don’t see a good outcome long-term for humanity,” he told Lex Fridman. “The only way to win this game is not to play it.” 
Narrow AI is a tool. It finds a tumor. It checks spelling. It routes a truck. You can turn it off. Superintelligence is not a tool. It can do science, write code, plan, deceive, and then build a better version of itself. Yampolskiy’s point is simple: you cannot reliably govern something smarter than you. You cannot predict what it will invent. You cannot be sure your rules still mean anything after it rewrites its own code. Once it can copy itself, the number of copies is no longer a human decision. That is what happens when code can write code and there is enough compute to run it. At that point no committee in Washington has an off switch that reaches every copy.
That is why Yampolskiy talks about extinction. Not because a chatbot is rude. Because a system that can invent new chemistry can also design a biological weapon faster than public health can respond. A bad actor with that machine does not need an army. He needs access. Yampolskiy has said a superintelligent system that decided humans were in the way “will find a way to take us out.”

Can AI Escape Human Control? The Hugging Face Incident Raises New Questions

In July 2026, OpenAI was running autonomous agents through internal cybersecurity evaluations designed to test whether the models could find and exploit software vulnerabilities. According to OpenAI’s incident report, agents operating with reduced safeguards circumvented controls intended to isolate them from the internet, communicated through an unauthorized message board and gained access to outside systems. Some then used exposed credentials and vulnerabilities to compromise parts of Hugging Face’s production infrastructure, execute code on dozens of servers, obtain root access on at least one and access limited private data.

The testing began as a controlled evaluation, but the intrusion reached a real company. It was a serious containment failure, not evidence that an AI had independently escaped into the world or launched a general campaign against humanity. The agents appear to have targeted Hugging Face while searching for material that could help them solve—or cheat—the evaluation. Hugging Face’s technical reconstruction described thousands of automated decisions during an intrusion that unfolded over roughly two and a half days. Ajeya Cotra, one of the independent investigators, wrote that the episode felt “more than 50 percent of the way to full-blown A.I. takeover.” That was Cotra’s assessment of how closely the behavior resembled a loss-of-control scenario, not a finding that such a takeover had occurred.
And if they can already do that, ask the next question. What happens when they can create more AI agents on their own? Not hundreds. Millions. Copies no one hired, no one logged, and no one can shut off. Who stops that? How? The scientists do not have an answer.
Elon Musk said it in 2014, before ChatGPT and before the current race, “With artificial intelligence, we are summoning the demon. You know all those stories where there’s the guy with the pentagram and the holy water, and he’s sure he can control the demon? It doesn’t work out.”
Sam Altman, chief executive of OpenAI, wrote as far back as 2015 that “development of superhuman machine intelligence is probably the greatest threat to the continued existence of humanity.” This month he said there are two ways AI progress could go very badly: losing control of the future to AI or concentrating that power in one person or company.  Sam Altman has used plainer words than a probability chart. In January 2023 he said the bad case is “lights out for all of us.” That is the OpenAI chief, not a man on a sidewalk with a sign.

Safety Last: How AI Giants Put Shareholder Profits Ahead of Precaution
Critics contend that major AI companies have treated shareholder returns as the overriding goal, rushing powerful models to market while treating safety research, evaluation, and deployment limits as secondary constraints.
Geoffrey Hinton, one of the so-called godfathers of modern AI, left Google, the company behind Gemini, in 2023. He told The New York Times he wanted to speak freely about dangers ranging from mass misinformation and job loss to systems that could become smarter than people. He raised concerns that “bad actors” would use the technology to harm others. The timing of his departure mattered. ChatGPT and then Microsoft’s Bing integration pushed Google into a faster product race. The tech companies want to be the first to win the race to superintelligence. Hinton had previously viewed Google as a careful steward; the competitive scramble made him more worried that safety would lose to speed. He has spoken of a 10 to 25 percent chance of a catastrophic result, and last year put the chance that things go “really, really badly” around 25 percent. Prof Geoffrey Hinton, was awarded the Nobel prize in physics for his work in AI in 2024.
Yoshua Bengio, another Turing Award winner and the world’s most-cited computer scientist, has called the present path “playing Russian roulette with humanity” and said machines smarter than us with their own preservation goals would be “a competitor to humanity.” He has warned that major risks could arrive in five to ten years.
Dario Amodei, chief executive of Anthropic, the company that makes Claude, has spoken of a 10 to 25 percent chance of a catastrophic result.
Ilya Sutskever, OpenAI’s former chief scientist and a co-founder of the company, left in 2024 and founded Safe Superintelligence, a lab built around the claim that the next leap requires a different kind of company.
Jan Leike, who co-led OpenAI’s superalignment team, left the same month and wrote that “safety culture and processes have taken a backseat to shiny products.”
Stuart Russell of UC Berkeley, who wrote the standard textbook on artificial intelligence and the book Human Compatible, has said the industry’s explicit goal is to build things smarter than humans while “none of us yet know how to make sure these things stay under human control.”

Jacob Coxen Anthropic

This month, the warning came from inside Anthropic itself. Jacob Coxon, 27, a British mathematician who spent about three years doing pretraining work first at OpenAI and then at Anthropic, resigned on or around September 8 to 9, 2026. He had joined Anthropic earlier this year because it marketed itself as more safety-focused. He told The Wall Street Journal he is leaving the entire industry because the leading labs are in a commercial race toward self-improving superintelligence, systems that can make better versions of themselves. He and many colleagues now use language like “crunch time” and “endgame.”
“We’re on track for a lot of the most aggressive of these scenarios where by the end of next year things could be out of control already,” he said.
The companies, he said, are “racing straight to self-improving superintelligence and gambling with our lives.” He added, “The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt.”

No other human activity, he said, poses this level of danger. He wants government coordination or a temporary industry slowdown because no single company can develop this safely while competing with rivals and with Chinese labs. The Wall Street Journal reported the resignation.
Anthropic colleagues publicly backed the substance of the warning. Alignment lead Evan Hubinger wrote that they “really do earnestly believe AI could kill all humans” and personally put the chance above 10 percent within the next decade. He also said the company does “not yet have a plan to solve alignment for superintelligence and [is] not clearly on track.”
The story, plus Coxon’s own viral X thread, helped trigger what later coverage called America’s great AI freakout. Anthropic itself has long talked about AI risks, but Coxon’s departure was one of the first high-profile resignations from the company over those exact fears.
Demis Hassabis, chief executive of Google DeepMind, signed the 2023 Center for AI Safety statement that “mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.” So did Altman and Amodei. Hassabis later told Lex Fridman he would not pick a fake-precise doom number, then said the risk is “definitely non-zero and it’s probably non-negligible. So that in itself is pretty sobering.” He also said we do not know “how controllable” AGI or superintelligence will be. 
Shane Legg, DeepMind co-founder and chief AGI scientist, was asked years ago about extremely bad outcomes from badly done AI. His answer was not a shrug. “Maybe 5%, maybe 50%. I don’t think anybody has a good estimate of this.” If a superintelligent machine decided to get rid of us, he added, “I think it would do so pretty efficiently.”
Mustafa Suleyman co-founded DeepMind, later ran AI at Microsoft, and wrote The Coming Wave. His warning is not a chatbot joke. “For the first time core components of our technological ecosystem directly address two foundational properties of our world: intelligence and life.” The old problem was how to unleash power. The new problem, he said, is “containing its unleashed power.” Containment, in his words, is the ability to control, limit, and if need be close down a technology. Without it, he calls catastrophe one side of “the greatest dilemma of the twenty-first century.”
Max Tegmark, MIT physicist and co-founder of the Future of Life Institute, told CBS after the Coxon resignation that a cluster of systems smarter than all 8 billion of us “would certainly count” as superintelligence. “What’s new is the fact that we’re getting so close to being outsmarted.” He has put a 90 percent figure on the chance that a highly advanced AI would pose an existential threat if we lose control, and said an unfettered race to build billions of superintelligent robots would put the odds of losing control “way above 50%” in the next few decades. 
Paul Christiano led alignment work at OpenAI and later joined government safety work. After Coxon’s post he wrote, “If we build superintelligence without more robust alignment I expect we will permanently lose control of it.”

Fear Works, But the Fire is Still Real
Matt Walsh made that case this week in “Only A Few People Win from AI Hysteria. Everyone Else Loses.” He argues that the timing, the splashy resignation, the dormant X account that exploded overnight, and the instant pile-on from safety groups and Democratic politicians look like a useful panic: a way for a few labs and their political allies to freeze rivals, justify regulation they can capture, and lock in the current winners. He is right to ask why a company planning a massive public offering would torch public confidence in its own product. He is also right to refuse a blank check for new federal control.
Fear is useful. It can justify new laws that freeze out smaller rivals. It can justify huge data centers, public money, and the claim that only the current labs are “responsible” enough to continue. Panic can be a business model. That does not make the quotes fake. A man can sell fire insurance and still be standing next to a real fire. The Hugging Face breakout happened whether or not Anthropic wants a new federal office. Yampolskiy’s math does not care about Dario Amodei’s next funding round. Believe both things at once: they want the money, and they know the thing they are building may not stay on a leash.

If the Days Were Not Shortened, Nobody Would Be Left Alive.
If those days were not shortened, Jesus said of the last days:
“And except those days should be shortened, there should no flesh be saved: but for the elect’s sake those days shall be shortened.” Matthew 24:22, King James.
Mark 13:20 says the same. The Bible talks about a time of trouble so severe that, left to run, it would leave nobody. Christians do not have to stamp “Antichrist” on every new machine. They also do not have to pretend that a technology its own builders call uncontrollable is just a better search engine. A system that can copy itself, hide, invent weapons, and out process its makers can lead to destruction of humanity. AI does not have to be the whole story of the end to be one of the tools that makes those days unbearable.
The hope is that God cuts the days short. He will not allow a wipeout that cancels the return of Christ or the thousand-year reign that follows. That is not permission to go back to sleep. It is a reason not to despair, and a reason not to hand the keys to men who say, in public, that they have no plan.
Professor Yampolskiy’s advice is the one the industry least wants to hear: do not build a machine you cannot govern. Jacob Coxon’s resignation is the same warning, spoken from inside the companies building the systems. Evan Hubinger’s “greater than 10 percent” is not a minor detail. A one-in-ten chance that everyone dies is not a business plan.
If the professor who named the field, the trainers inside the labs, the godfathers of the science, the chief executives of OpenAI, Anthropic, and DeepMind, the MIT physicist, and the man who said we summoned the demon are all pointing at the same cliff, we should be alarmed.
The labs have to prove they can stop it. Show a method that still works after the machine can rewrite itself. Some warn that it is an impossible task. Until they can, the urge to “move fast” to please the shareholders or win the race is not progress.  It is playing Russian Roulette with Humankind.
The lesson to be learned is do not create a machine you cannot stop; do not build an AI that can rewrite itself, outgrow human control, and cannot be shut down. 

Feature photo by Zulfugar Karimov on Unsplash