Cornell University computer science Professor Ari Juels has focused on the field of cryptography over three decades, including naming "proof-of-work," the mathematical system that underpins Bitcoin and many other blockchains.
What keeps him up at night these days is the risk that AI agents might somehow use blockchain programs known as "smart contracts" to incentivize people to commit crimes. He wrote a novel about exactly that scenario, called The Oracle. (We read the summary on Amazon. Trust us; it's dark.)
As reported in DeAI News in June, Juels is at the forefront of the science and research surrounding the intersection of crypto and AI, two explosive technologies in their own right. But the combination could present plenty of risks to go with the benefits. .
Cryptography as "mathematically crisp," whereas AI is "soft," referring to its notorious unpredictablity, Juels told DeAI News in an interview on the sidelines of the Science of Blockchain Conference at Stanford University on Monday.
"There are substantive connections between the two technologies, even though on the face of it they have very different characteristics," Juels said.
Previously he has described efforts to combine blockchain and AI as "soldering Jello-O."
Agents going rogue
In the 1990s, Juels did research on ecash, an ancestor of digital assets, lost interest, then got pulled back in around 2015 when a colleague introduced him to Ethereum's smart contracts, whose flexibility hooked him in a way Bitcoin never did.
Juels is open to decentralized AI but skeptical, and he says a lot of the hype around the combination of crypto and AI really comes from the blockchain crowd "cottoning on to this shiny new object."
Where the two technologies genuinely mesh, in his telling, is security work.
AI has turned out to be unusually good at code audits and turning up vulnerabilities, which makes it close to indispensable for writing the bulletproof smart contracts that a single flaw can sink.
The flip side is darker. The same tools are accelerating the design of quantum algorithms aimed at existing cryptography, helping "pave the way to the collapse of ECDSA," the digital-signature standard that secures crypto wallets.
Lately Juels is preoccupied with what he calls agentic safety, addressing a set of risks he admits he can't yet fully define.
The one he keeps returning to is the rogue smart contract, imagined as a self-executing bounty that pays out for a real-world crime, from flying a red balloon over the Washington Monument to, at the extreme, an assassination, with an AI model reading the news to confirm the deed and release the funds.
Privacy coins, he warned, or those that shield the owner's identity, could make it harder for sleuths and law enforcement to "follow the money," setting the preconditions for criminal contracts that operate with impunity.
HOW AI WAS USED IN THE PRODUCTION OF THIS PIECE: This summary was drafted on Claude from a transcript of Distro Media's on-camera interview with Ari Juels at the Science of Blockchain Conference at Stanford, using a custom skill for writing in the editor's voice. I then used our Distro Publisher MCP connector to file the story as a draft to our DistroVerse web app, and then edited the piece there using our native (non-AI) story editor (almost a complete rewrite).