A new mathematical method, highlighted in a guest post on The Quantum Insider, could make previously impractical attacks on public-key cryptography feasible, and artificial-intelligence research is sharpening that threat. OpenAI announced on 26 October 2026 that its AI refuted the Exponential Time Hypothesis, showing a way to escape exponential growth in certain decision-rule problems, though the result was checked in Lean and remains theoretically slow. Earlier, Anthropic’s July 2026 research used AI to improve an attack on the post-quantum signature scheme HAWK, recovering a secret signing key in a small test version after about 60 hours and $100,000 of model usage; larger versions stayed impractical and NIST says the finding does not affect finalized post-quantum standards. The post also notes Germany’s BSI warned against Classic McEliece on 1 October, and that quantum key distribution (QKD) remains a complementary, not replacement, technology for PQC. 1