AI Advances in Mathematics Spark Concerns Over Global Encryption Security

Advances in AI mathematics threaten encryption methods that protect digital information, prompting efforts to verify both widely used systems and planned next-generation safeguards. Regulators, including the UK's Financial Conduct Authority (FCA), are increasingly urging companies to prepare to change the encryption methods protecting sensitive online information. Experts said AI-assisted mathematical breakthroughs raise questions about whether AI could soon defeat core cryptographic methods protecting online bank accounts and databases. However, AI has not yet broken a core cryptographic method, and some cyber security experts doubt AI will do so.

Calls for preparations intensified after OpenAI sparked controversy by releasing hundreds of mathematics papers, including purported proofs addressing longstanding problems. OpenAI said some solutions in the papers were generated by OpenAI's AI models. The Association for Human Mathematics, established to defend mathematics against threats from AI, criticised OpenAI's work and urged mathematicians to end collaboration with OpenAI. OpenAI defended the company's actions and pledged $5 million to a foundation focused on mathematics.

Australian mathematician Geordie Williamson warned during a recent Imperial College London lecture that AI systems were changing which problems count as difficult. Williamson said mathematical difficulty underpins much modern cryptography and warned of major, unsettling shocks ahead. Improvements in AI capabilities in algorithm design and number theory, which studies whole-number properties, have increased the ability to solve previously difficult problems. A widely used cryptographic approach relies on multiplying two very large prime numbers. Existing computers can readily perform that multiplication, but identifying the original pair from the result is much harder.

AI is already affecting cyber security. Companies worldwide rushed last week to repair vulnerabilities in PHP, a widely used language for programming websites. AI models from OpenAI and Google helped identify those vulnerabilities.

AI's mathematical progress has challenged the assumption that quantum computers pose the most urgent threat to cryptography. Quantum computers are designed to use subatomic physics to produce results beyond the capabilities of classical computers. However, most experts believe a quantum computer powerful enough for this purpose remains years away. A finance industry insider questioned whether AI or quantum computing posed the nearer threat to cryptographic algorithms.

The FCA said cyber threats are changing rapidly as AI and quantum computing progress. The FCA said the regulator is testing a new supervisory tool with 50 large institutions to assess readiness for emerging technological threats to cryptography. The FCA plans to issue practical guidance next month covering AI and quantum-computing risks for the more than 35,000 financial services companies the regulator supervises in the UK. The FCA urged firms to act immediately, understand existing system protections and begin preparing for quantum-secure systems.

Daniel Shiu, a cryptomathematics professor at the University of Manchester, described AI's potential to defeat mathematically based forms of cryptography as a significant risk. Shiu said potential vulnerabilities include some less-developed algorithms within a new suite intended to withstand quantum-computer attacks. Shiu said AI's growing mathematical capabilities were prompting authorities to act. Shiu said countries including the US, China and Germany had reassessed existing and proposed cryptographic tools because of those capabilities.

Telecommunications group BT said the company was already preparing to adopt new encryption methods because of future quantum-computing threats. BT said the company was also working to make cryptographic technologies easier to change as standards, technologies and threats develop.