For 35 years, a cryptic sculpture at CIA headquarters has stumped the world's top code breakers. Now, AI chatbots are generating a flood of wrong answers from overconfident amateurs, much to the creator's dismay.
Artist Jim Sanborn, who designed the Kryptos sculpture in 1990, is getting bombarded with emails from people claiming they've cracked its final unsolved cipher using AI tools like Grok 3. While three of the sculpture's four encrypted panels were solved in the 1990s by the CIA, NSA and others, the last segment known as K4 remains unbroken despite decades of effort by expert cryptographers.
"The numbers [of submissions] have increased dramatically," Sanborn told Steven Levy. "The people that did their code crack with AI are totally convinced that they cracked Kryptos during breakfast!" The 80-year-old artist is frustrated not just by the volume of incorrect solutions, but by how AI has changed the nature of the attempts. Where dedicated code breakers once approached the puzzle with careful study, now people rely on chatbots that confidently generate wrong answers.
Over the years, Sanborn has released several clues to help solvers, including that the plaintext contains the words "BERLIN," "CLOCK," and "NORTHEAST." But even these hints haven't led to a breakthrough. He plans to take the solution to his grave, possibly leaving his wife to decide its fate.
"AI lies, and does not have enough info," Sanborn told one misguided solver. The artist sees the flood of AI-generated attempts as counter to the sculpture's artistic purpose — a meditation on the nature of secrecy and intelligence gathering.
Previously:
• 'Kryptos' sculptor teases clues to C.I.A. cryptosculpture
• NSA cracked CIA 'Kryptos' sculpture before CIA
• Cracking Kryptos
• Can you solve the 'secret' phrase on the USPS's new Mystery Message stamps?
