Claude AI Revisits Mt. Gox Code — And Finds Every 2011 Vulnerability
What if artificial intelligence had existed during Bitcoin’s first great collapse? Former Mt. Gox CEO Mark Karpelès decided to find out — by feeding the exchange’s old code into Claude AI from Anthropic.
A Test in Digital Archaeology
Over a decade after the infamous hack that shattered the Mt. Gox empire, Karpelès ran an experiment no developer of the 2010s could have imagined.He uploaded the exchange’s 2011 source code to Claude AI and asked it to audit the system for security flaws.
The AI didn’t hesitate. After analyzing the legacy PHP and Python files, it labeled Mt. Gox as a “functional but critically unsafe Bitcoin exchange.”
Karpelès also provided GitHub history, server logs and data once leaked by hackers — a complete time capsule of the exchange’s infrastructure.
AI Finds What Humans Missed
Claude’s report reconstructed the original weak points that led to the collapse: unvalidated inputs, poor key management and a single administrator with too much control. It also flagged a long-forgotten database bug that allowed double-spending in certain edge cases.In a surprising twist, Claude praised the exchange’s founder, Jed McCaleb, calling his three-month build “architecturally ambitious and efficient for its time.” Yet the AI noted that the system lacked segregated wallets and modern encryption practices — flaws that would prove fatal by 2014.
Could AI Have Prevented Mt. Gox?
Karpelès reflected on how different the story might have been if AI-driven code review had existed back then.He called Claude’s audit “a look into the past with tools from the future,” adding that such technology could one day make crypto exchanges virtually self-auditing.
Security experts agree that AI now plays a vital role in detecting patterns of exploitation and preventing systemic failures. Modern LLMs are already being used by auditing firms to examine smart contracts and automate compliance checks.
A Lesson from History
For the crypto industry, Mt. Gox remains a symbol of how fast innovation can outrun security. Karpelès’s experiment suggests a way to close that gap — by letting AI help guard the systems it once could only watch fail.If Claude had been around in 2011, perhaps the world’s first Bitcoin exchange would still stand today — and the phrase “Mt. Gox collapse” would never have entered crypto folklore.
Editorial Team — CoinBotLab