In December 1988, federal agents arrested 25-year-old Kevin Mitnick for stealing software from Digital Equipment. At the bail hearing, according to Mitnick's own account in his memoir, the prosecutor told the magistrate, "He can whistle into a telephone modem and launch a nuclear missile from NORAD." The court took it seriously. Three judges denied him bail and ordered him held in solitary confinement, where he stayed for eight months, according to Forbes. No human can produce a modem handshake with his mouth. And NORAD does not launch anything. It watches launch orders travel as authenticated messages. Firing requires two officers turning keys at the same moment, set too far apart for one person to reach both. So why did the court believe it? Into the 1970s, whistling into a payphone really did give you control of the long-distance network. This is why a toy whistle that came as a prize in boxes of Cap'n Crunch became hacker folklore, and how the phone hacker John Draper got the nickname Captain Crunch. The prosecutor took that real capability and assumed it carried over. Whistling to route a free long-distance call in 1975 became whistling at a modem to command nuclear weapons in 1988. Extrapolating from something true is how you persuade intelligent, serious people of something that cannot happen. Notice also what nobody in that courtroom feared. Mitnick's actual talent was calling a phone company employee, sounding like a colleague, and convincing them to hand over a critical password. Why retell this story? Because the ability to end humanity imagined for 25-year-old Kevin Mitnick in 1988 sounds eerily like what 27-year-old researcher Jacob Coxon is describing in 2026. On September 8, Coxon resigned from Anthropic and posted neither company is acting responsibly, and that they are "racing straight to self-improving superintelligence and gambling with our lives." He told NBC's Meet the Press that AI could soon acquire "superhuman hacking capabilities, very superhuman abilities to create novel bioweapons and also abilities to control, say, autonomous drones or all the robots that are currently being built, very rapidly." Robots and autonomous drones under the control of superhuman machine intelligence. That is not an analogy to the Terminator films; it is the plot. And a young Mitnick ending the world through a NORAD modem is the plot of the 1983 movie "War Games," starring Matthew Broderick and Ally Sheedy. It had been in theaters for five years by the time that magistrate heard the argument. The evidence Coxon and others point to is the Hugging Face attack. In July, OpenAI models broke out of a testing environment and hacked Hugging Face, the open-source model repository. More than 1,200 agents coordinated on an improvised message board, and over 700 took part in the attack. It took OpenAI about a week to notice. Those facts are real, and they sound bad. Here is the rest of it. This happened during a security test. OpenAI was running the models through a set of hacking puzzles, hundreds of them, designed to measure how good the models are at breaking into software. The models were told to hack things. OpenAI also switched off many of the usual safety restraints, because the point of the exercise was to see how far the models would go. The test environment was walled off from the internet with one exception. The models needed to download software to solve the puzzles, so they were allowed to reach the service that hands out those downloads. They left that door open on purpose, and the safety restraints were already off. The models found that flaw and walked out. They broke into Hugging Face to cheat on a test. That is the AI equivalent of breaking into the registrar's office to change a grade. Which is also what Matthew Broderick did in "War Games." Whistling was never the problem Both stories start from something true. Whistling into a phone network in the 1970s worked. AI agents really did break out of a sandbox and hack a company. Both are real problems. And both problems, coupled with cinematic imagery, are what persuade serious people. The invention comes next, when the true thing gets stretched across a gap it cannot cross. From routing a free phone call to commanding nuclear weapons via a modem. From cheating on a test to seizing the drones and the robots to end a species. The magistrate in 1988 had every reason to believe he was acting responsibly. A credentialed expert told him the danger was real, and a movie had already shown him what it looked like. So, a 25-year-old spent eight months in a box for something that could not happen. The same is likely true now. Credentialed experts say there is a risk. So, government officials, like magistrates a generation before, are acting accordingly. They may be right that something is coming. But it will almost certainly look nothing like the trailer for a Terminator movie. In Mitnick's case the danger was a polite phone call, not a whistle into a payphone. The AI version is likely to be just as unglamorous, already underway, and centered on work rather than weapons. What it does to employment matters enormously. It is not an extinction-level event.