The founder of Internet Computer (@dfinity, $ICP) says artificial intelligence, rather than quantum computing, is the more pressing risk to crypto security. Dominic Williams argues that recen
The founder of Internet Computer (@dfinity, $ICP) says artificial intelligence, rather than quantum computing, is the more pressing risk to crypto security. Dominic Williams argues that recent AI-driven results show how quickly assumptions in cryptography and computer science can change.
Williams points to a claim that AI has disproved the Exponential Time Hypothesis (ETH), a long-standing conjecture in computational complexity. OpenAI said its system proved the hypothesis false. The hypothesis concerns the complexity of Boolean formula satisfiability and has been used in hundreds of papers to derive hardness results.
The result is not yet settled. None of the three results OpenAI announced has been confirmed by independent verification.Both OpenAI and Anthropic say their refutations were verified in Lean, a formal proof language, and OpenAI has published its raw proof on GitHub for outside experts to examine.
The claim also does not mean current cryptography is broken. Williams said OpenAI checked whether its result weakened cryptographic systems and was "reasonably confident" nothing was newly insecure. (This is a different Williams, the MIT computer scientist Ryan Williams, who is among the outside reviewers.) The algorithm improves solving efficiency for Boolean formulas but does not break ECDSA or Schnorr signatures. That matches the view of ICP's founder that core algorithms are not in danger for now, even though the episode shows AI could change the playing field for crypto.
The ICP founder's criticism is aimed at blockchains adopting "quantum-safe" signatures. He argues that practical quantum computing is still years away, so real usage of these schemes is likely to be very limited, while newly designed and complex algorithms can carry serious bugs.
The HAWK signature scheme offers a recent example. On July 28, 2026, Anthropic announced it had used an AI model to find a vulnerability in HAWK, a lattice-based signature algorithm under consideration for standardization. The HAWK team then withdrew it, and it will not be standardized or deployed.The attack was discovered primarily by Anthropic's Claude Mythos Preview model with minimal human guidance.
The impact is limited. NIST says the finding does not affect its finalized post-quantum standards, such as ML-KEM and ML-DSA, which rely on different mathematical foundations.HAWK was not a final standard and was not deployed in production, so no emergency remediation was needed.
Williams says ICP will not take part in what he calls the "post-quantum mania." The debate reflects a wider split in the industry: NIST continues to urge migration to its finalized standards, while ICP's founder argues that rushing into newer schemes brings its own risks, particularly as AI tools get better at finding flaws.
Sources
