Whether AI is weakening the cryptography we rely on: attacks and cryptanalysis found or sped up by AI, AI mathematics with a bearing on lattices, elliptic curves, hashes and number theory, parameters revised and schemes dropped because of it, and what cryptographers and protocol designers say about the threat. Topics and rules ↓
Yehuda Lindell rejected Justin Drake's claims that AI could break elliptic curve cryptography, stating the predictions lack concrete evidence. Lindell noted breaking ECC would have catastrophic consequences across finance but found no logical basis for the warnings.
OpenAI released over 700 mathematics paper manuscripts on October 13, 2026, prompting Ethereum researcher Justin Drake to warn that AI could break ECDSA within months. ECDSA secures blockchain wallets, financial systems, encrypted communications, and IoT devices; experts including UT Austin's Scott Aaronson disagree on whether evidence supports the claim.
Ethereum co-founder Vitalik Buterin posted that AI-accelerated mathematics could substantially weaken lattice-based cryptography (ML-DSA, FHE) within two years, similar to how the number field sieve degraded RSA. He recommends hash-based signatures (WOTS, SPHINCS), 10× larger lattice parameters, and off-chain encrypted notes.
Cryptography researcher Matthew Green used Claude and GPT 6 Astra to survey historical cryptanalysis results and estimate human-hours spent attacking public-key encryption schemes. The work aims to inform ongoing debates about AI threats to cryptography.
Dan Boneh, Eddy Lazzarin and others examined whether recent AI math breakthroughs threaten elliptic curves or lattice assumptions, finding no cryptographic vulnerabilities demonstrated. The assessment separates real risk from speculation following AI-solved math problems.
Johns Hopkins cryptographer Matthew Green estimates 1% odds public-key encryption is mathematically broken and 15% odds the world functionally loses confidence in it, arguing AI could surface cryptanalytic surprises faster than standards bodies can respond. He cited an AI-found attack that lowered HAWK-512's security from 2^150 to 2^108, leading NIST to withdraw it.
Dan Boneh, Eddy Lazzarin and other cryptographers from Stanford and a16z examined whether recent AI breakthroughs in mathematics threaten public-key cryptography, finding no demonstrated vulnerabilities in elliptic curves or lattice-based schemes. They noted that while AI has solved hard math problems, the published results contain no cryptographic attacks. The experts separated real risk from speculation: modern cryptography rests on the assumption that certain problems have no efficient solution, and no AI, however capable, can find an algorithm that does not exist. Their assessment clarifies that AI progress in mathematics does not automatically translate to breaks in deployed encryption.