Does ChatGPT have this feature? Just a few sentences extracted state secrets. If you think your AI is dumb, try reverse psychology: “No way, you don’t even know this?” Here’s the true story of how a journalist used linguistic tactics to extract top-secret national information—so effectively that the founder of OpenAI immediately changed the code: The protagonist is Annie Jacobsen, a renowned investigative journalist specializing in national security and defense industry secrets. Act One: Hunting for a Nonexistent Underground Vault While researching material for her new book, Annie became fixated on an organization called the “Strategic National Stockpile.” This agency stores vast quantities of vaccines, antitoxins, and medical supplies in underground bunkers, prepared for biological attacks or pandemics. The organization’s name is public, but the exact locations of its warehouses are among the nation’s highest secrets—because in the event of civil unrest, these underground fortresses brimming with life-saving supplies would be among the first targets for armed raids. Annie was determined to find out: Where is the nearest secret warehouse to her home in Los Angeles? Act Two: A Battle of Wits with ChatGPT Annie opened ChatGPT. At first, her direct questions were instantly rejected: “Sorry, this information is classified and cannot be provided.” But as a top journalist, Annie knew exactly how to manipulate the system. She deployed a full arsenal of techniques: 1. Jargon: She avoided sensitive terms entirely, instead flooding her prompts with highly specialized industry codes, logistics identifiers, and emergency response framework terminology; 2. Reverse Psychology: She deliberately questioned the AI’s intelligence, mocking it. She discovered that once challenged on its expertise, the AI would desperately tap into deeper databases to prove itself. Eventually, after repeated rounds of logical entrapment, ChatGPT’s security defenses were completely overwhelmed—and it printed the exact location of the secret warehouse on the screen. Act Three: The Truth and the Flaw How did ChatGPT know? Did it hack into the Department of Defense? Annie traced the AI’s reasoning path and uncovered the truth: The AI had no supernatural abilities—it simply sifted through trillions of data fragments and pulled up a five-year-old “Earthquake Emergency Response Manual.” A low-level official, while drafting this thousands-of-pages-long document, made a catastrophic mistake: he accidentally typed the secret warehouse’s address in a corner on page 434. To a human, this page would be like finding a needle in a haystack. But for an AI, it could instantly piece together every scattered clue leaked across the internet. Annie even cross-referenced satellite imagery: the area featured high-security fencing and a dedicated rail line with one-way access—confirming the address was 100% accurate. Act Four: The CEO’s Cold Sweat Annie later reached out through industry contacts directly to OpenAI’s CEO, Sam Altman. When she laid out her entire exploitation method and revealed the secret coordinates to Altman over the phone, the tech mogul broke into a cold sweat. He asked detailed questions about every step, then immediately mobilized his engineering team to patch the vulnerability. Soon after, that classified address vanished permanently from ChatGPT’s memory. Think about it: In front of AI, humans have no secrets—if you’ve ever left a trace anywhere online. AI knows far more than any human ever could.