AI maths advances raise fears over security of global encryption
Rapid AI progress in mathematics is raising concern that AI could eventually crack the core cryptography behind global encryption. OpenAI recently unveiled hundreds of mathematical papers, including 'proofs' of longstanding problems it said included AI-generated solutions; the Association for Human Mathematics attacked the release and urged mathematicians to stop working with OpenAI, and OpenAI pledged $5mn to a maths-focused foundation. Mathematician Geordie Williamson describes 'a complete kind of restructuring of what difficulty is' and expects 'very unsettling major, major shocks'. Manchester's Daniel Shiu calls AI a 'significant risk' for 'more mathematical forms of cryptography', including 'less mature' post-quantum algorithms, and says this has prompted re-evaluation in the US, China and Germany. The FCA is trialling a supervisory tool with 50 large institutions and will issue guidance to more than 35,000 firms next month ('Firms shouldn't wait to act'); BT is preparing to migrate to new encryption methods. Last week companies rushed to fix PHP flaws found partly by OpenAI and Google models. Counterweights: AI has not yet broken any core cryptographic method; some cyber experts are sceptical; much of the piece is expert speculation; and most experts think a powerful quantum computer is still years away.
Why it matters
If AI, not quantum computing, becomes the nearer threat to encryption, the timeline and scope of the post-quantum migration change, including for 'less mature' post-quantum schemes. That matters for cryptography and security start-ups and for every regulated firm the FCA is about to address. No break has been shown, so this is a risk signal, not an event.
