AUTO-UPDATED

Can AI break encryption? OpenAI’s new math has cryptographers divided

OpenAI recently released 722 mathematical manuscripts generated by artificial intelligence, sparking expert debate over whether advanced models could eventually compromise the encryption systems securing global digital communications.

Key Points

  • OpenAI published 722 AI-generated math papers, though three were retracted shortly after release due to inaccuracies.
  • Cryptographers like Matthew Green warn that AI could potentially identify mathematical shortcuts, threatening the security of public-key encryption algorithms.
  • Research from ETH Zurich and Anthropic shows AI models are rapidly improving at cryptanalysis, including finding weaknesses in quantum-resistant schemes like HAWK.
  • Experts like Yehuda Lindell and Bart Preneel argue there is currently no evidence that AI has broken any widely deployed cryptographic standards.
  • Security professionals emphasize that stealing encryption keys remains a more immediate and practical threat than the theoretical mathematical collapse of encryption.

Why it Matters

Public-key cryptography serves as the foundation for secure online commerce, banking, and private messaging. If AI models successfully discover fundamental mathematical flaws, it could force a massive, high-risk global infrastructure upgrade to protect sensitive digital data.
Scientific American Published by Chris Stokel-Walker
Read original