Cryptographers split over whether AI math breakthroughs could break encryption

The latest:

The threat to internet encryption may arrive from artificial intelligence rather than quantum computers, The Wall Street Journal reported, citing cryptographers tracking a run of AI-generated mathematical results. Johns Hopkins cryptographer Matthew Green said recent advances could eventually touch the math underpinning encrypted communications. Coinbase cryptography head Yehuda Lindell rejected that framing, arguing speculation about an imminent collapse helps no one.

Details:

• The core assumption: Encryption rests on operations easy to perform in one direction and impractical to reverse, according to the report. Multiplying two large primes is simple; factoring the result back is effectively impossible on conventional computers. If that asymmetry weakens, digital locks weaken with it.
• Green’s argument: Green said cryptography’s security has rested largely on many smart people examining it for a long time without finding a break. He argued that standard no longer holds, telling the Journal that human beings looking at things “is not the gold standard—it’s not relevant anymore.”
• The disputed result: Green cited AI-lab breakthroughs including a partial solution to the Navier-Stokes equations and, more recently, a result disproving the Exponential Time Hypothesis. That hypothesis held that certain problems have no shortcuts; if false, decades of theoretical computer-science work could lose relevance.
• Buterin’s warning: Ethereum creator Vitalik Buterin posted on X in early October, before the latest OpenAI and Anthropic math results, urging that risks to cryptography from AI-accelerated math be taken seriously. On Monday he repeated the call at the Ethereum Foundation’s annual conference in Mumbai, pressing for post-AI cryptography.
• The rebuttal: Lindell, Coinbase’s head of cryptography and a Bar-Ilan University professor on leave, said anything is possible but speculating about bad outcomes helps no one. He objected specifically to claims that postquantum algorithms now being deployed are vulnerable to AI-driven mathematical discoveries.
• The fallback: Lindell assessed that evidence points the other way, with postquantum algorithms likely to withstand most attacks from any kind of computer. At worst, he said, engineers would lengthen cryptographic keys, slowing encoding and decoding online without rendering systems unusable.
• AI as defense: Lindell said AI will very likely assist cryptographic analysis and could help strengthen algorithms rather than only undermine them. His position rests on the assumption that genuinely hard problems with no shortcut exist, a premise he expects to survive.
• Who finds it first: Even in a worst case, Lindell said he would prefer a lab such as OpenAI or Anthropic make the discovery, citing their record of disclosing vulnerabilities for patching. He warned against a system being broken by North Korea without anyone else being told.
• The timing problem: Green countered that cryptographic systems take years to update, so discreet disclosure would not be enough protection. He said he has no idea what is coming.

Background:

Postquantum cryptography is a set of algorithms built over decades to survive attacks by quantum machines, which experts have long warned could unravel today’s encryption. The report compared the moment to Bletchley Park, where Alan Turing’s team broke Germany’s Enigma code many had considered unbreakable.

Between the lines:

The dispute is not about whether AI improves at mathematics — both sides accept it will. It turns on whether an undiscovered shortcut exists for problems assumed to have none, which is why the Exponential Time Hypothesis result carries weight beyond its field. Lindell’s fallback of longer keys and Green’s warning about multi-year update cycles describe the same gap: the fix is known, the time to deploy it is not.

What’s next

Watch for further mathematical results from OpenAI and Anthropic touching cryptographic foundations, the rollout pace of postquantum algorithms across major systems, and whether the Exponential Time Hypothesis disproof survives peer scrutiny.