AI Chatbots Spark Controversy in Attempts to Crack CIA's Kryptos Sculpture Code

AI Thinks It Cracked Kryptos. The Artist Behind It Says No Chance

The Kryptos sculpture located on the grounds of the CIA headquarters, the George Bush Center for Intelligence in Langley, Virginia.Courtesy of the Artist For 35 years, amateur and professional cryptographers have tried to crack the code on Kryptos, a majestic sculpture that sits behind CIA headquarters in Langley, Virginia. In the 1990s, the CIA, NSA, and a Rand Corporation computer scientist independently came up with translations for three of the sculpture's four panels of scrambled letters. But the final segment, known as K4, was encoded with knottier techniques and remains unsolved. This failure has only deepened the obsession of thousands of would-be cryptanalysts. When one of them thinks they have an answer, they write to Jim Sanborn for confirmation. Sanborn is the artist who created the installation and the only person who knows the answer. Lately the pace has picked up. And Sanborn is getting ticked off -- though not for the reasons you might think. Consider the email from one recent would-be codebreaker. "What took 35 years and even the NSA with all their resources could not do I was able to do in only 3 hours before I even had my morning coffee," it began, before the writer showed Sanborn what they believed to be the cosmically elusive solution. "History's rewritten," wrote the submitter. "no errors 100% cracked." You might ask, what enables someone to believe they'd outperformed the world's most elite mathematicians and cryptologists, including some spooks who maybe have a quantum computer in the basement? The answer is pure 2025: a chatbot! It turns out that the current generation of AI models is happy to accept prompts aimed at solving Kryptos, coming up with the decoded message in plaintext, and declaring victory. Sanborn says he's seeing it more and more. Of course, this writer's "solution" was dead wrong, like the thousands Sanborn had previously bounced. Sanborn contacted me recently to express his disgust with this development. "It feels like a major shift," he says. "The numbers [of submissions] have increased dramatically. And the character of the emails is different -- the people that did their code crack with AI are totally convinced that they cracked Kryptos during breakfast! AI seems to be lying to them, telling every one of them that it's 99.99% sure that they cracked Kryptos, congratulations. So they all are very convinced that by the time they reach me, they've cracked it." This bothers Sanborn in several ways. Until recently there was an unspoken agreement between the artist and the Kryptos faithful that the effort to crack the code would be taken seriously. (Some years ago, Sanborn began charging $50 to review solutions, providing a speed bump to filter out wild guesses and nut cases.) That back-and-forth fed into the artistic nature of Kryptos; having an object that defies solution in the backyard of the CIA is a subversive commentary on the funhouse-mirror aspect of intelligence gathering, where every truth is cast into doubt. The fact that thousands of people have spent an enormous amount of effort to unveil the plaintext -- which, judging from the decoded panels so far, indicates Sanborn's message is a gloss on secrecy itself. Newcomers seem to have no sense of this complexity. "The crowd of people trying to crack Kryptos today have no idea what Kryptos is," says Sanborn. He finds himself sifting through emails from randos using AI shortcuts that require little thought and expertise, let alone appreciation for the challenge. It's like saying you've scaled Everest by taking a helicopter ride to the summit -- but worse, because these ankle-biters haven't solved the code at all. They've barely climbed above sea level. Sometimes, in his replies, Sanborn doesn't hold back. "I infer from your certainty that you used AI," he told one misguided guesser. "AI lies, and does not have enough info."

Chatbots Convinced Idiots They Cracked the Code on a Sculpture in the CIA’s Backyard

Near the CIA headquarters in Langley, Virginia, there is a sculpture known as Kryptos. It has been there since 1990 and contains four secret codesâ€"three of which have been solved. The final one has gone 35 years without being decrypted. And, according to a report from Wired, the sculptor responsible wants everyone to know that you are not solving the damn thing with a chatbot. Jim Sanborn, who has created sculptures for the Massachusetts Institute of Technology and the National Oceanic and Atmospheric Administration in addition to his work outside CIA HQ, has reportedly been inundated with people who are absolutely positive they've solved K4, the final panel of unsolved code that he encoded with knottier techniques. But these code crackers aren't the actual cryptanalysts, professional or otherwise, who have been obsessed with decoding the message since it first appeared. No, these are just some people who ran the code through a chatbot and have taken its word on the answer. In a conversation with Wired, Sanborn said that he's seen a significant uptick in submissions, which is already annoying if you're already a 79-year-old who has gotten so many submissions over the last three in a half decades that you had to start charging $50 fee for reviewing solutions because you'd had so many cranks message you over the years. But worse than just the frequency of submissions, according to Sanborn, is the attitude of the submitters. "The character of the emails is differentâ€"the people that did their code crack with AI are totally convinced that they cracked Kryptos during breakfast," he told Wired. "So they all are very convinced that by the time they reach me, they've cracked it.†Here are just a few samples of the extremely arrogant and self-satisfied messages that Sanborn has received in recent years: "I’m just a vet...Cracked it in days with Grok 3.†"What took 35 years and even the NSA with all their resources could not do I was able to do in only 3 hours before I even had my morning coffee." "History’s rewritten...no errors 100% cracked.†If you've spent any amount of time on social media, particularly on X, you've seen these people. Maybe not these same guys, but the same type of guy. You know, the ones who say "Just Grok it" or reply to someone's post with "Here's what Grok says," or share a screenshot of ChatGPT's response as if that is in any way additive to the conversation. The smugness is, frankly, inexplicable. Even if they did successfully crack Sanborn's code using AI (which, for the record, Sanborn says they haven't even gotten close), what is it about asking a machine to do the work for you that generates such self-satisfaction? It'd be one thing if they trained a large-language model on an endless amount of encryption knowledge and used that to crack Sanborn's code. But they're literally just asking a chatbot to look at a picture and solve it. It's the least clever thing imaginable. It's flipping to the back of the textbook to see what the correct solution to the math equation is, except in this case, the textbook has confidently hallucinated the answer. This behavior isn't uncommon, really. A study published last year in the journal Computers in Human Behavior found that when people learn advice was generated by AI, they tend to over-rely on it, even allowing it to convince them to go against conflicting contextual information and their own personal interests. The same study found that when someone over-relies on AI advice, it negatively affects their interactions with other humans. Maybe it's because they're so damn pleased with themselves.