Abstract:
Last Tuesday, OpenAI announced that its internal models had overcome a number of unsolved mathematical problems. This news caused concern among the world's top cryptographers. Cryptozoologists specialize in information encoding techniques to keep information private and secure. They worry that as AI continues to make progress in the field of mathematics, it may be able to find new algorithms in the future to break through various encrypted services - from online payments to instant messaging platforms such as WhatsApp.
The underlying logic of modern Internet security is based on the premise that some mathematical operations are easy to perform, such as using public keys to encode information; but if the specific value of the private key is not known, reverse operations and decoding will be extremely difficult. This set of assumptions of "forward is easy, reverse is extremely difficult" is the foundation of secure communication on the Internet, supporting e-commerce, cryptocurrency and various digital account systems.
AI models from OpenAI and other organizations have successively solved problems that have plagued the mathematical community for decades. This brings up the possibility that a powerful AI system may be able to bypass the private key and directly crack the encryption problem.
Whitfield Diffie is the co-inventor of the Diffie-Hellman key exchange algorithm in 1976 and one of the founders of public-key cryptography. "I expect that the field of mathematics will usher in huge changes, and the field of cryptanalysis will not escape this change," he told this magazine in an email. "But I can't say the time scale. One year may be too early, and five years may not be enough." Diffie currently works at the Stanford Center for International Security and Cooperation as a visiting researcher.
Martin Herman, an emeritus professor of electrical engineering at Stanford University who pioneered public-key encryption technology with Diffie, is relatively optimistic. He believes that in the short term, AI-assisted attacks will have an impact on network security; but in the long term, humans will use AI better as a tool to patch various security vulnerabilities.
In contrast, some of the remarks made on the X social platform last week were more sensational. Matthew Green, professor of computer science at Johns Hopkins University and founder of encryption start-up Zeutro, wrote after OpenAI released the mathematical results: "I think the public key cryptography system may become invalid."
Justin Drake, a security researcher at the Ethereum Foundation, called on the blockchain industry to prepare for AI's possible breakthrough in the field of cryptography.
OpenAI’s related actions in the field of cryptography
What exacerbates the unease in the industry is that the large number of mathematical research results released by OpenAI
do not contain any cryptography-related conclusions
. Many experts have speculated that OpenAI has major achievements in cryptography, but has chosen not to disclose them to the public.In an interview with "The Information", Green said that due to national security considerations, OpenAI may first disclose cryptography-related discoveries to the US government. Hed Kowitz, CEO of cybersecurity company Silverfort, said the public may never learn about the results of such research.
OpenAI did not respond to questions about whether its internal models were used to solve cryptographic problems.
Green believes that major AI companies have invested a lot of resources in studying cryptography issues. In fact, OpenAI released ten mathematical advances derived from its latest model Astra in September, covering quantum complexity and lattice cryptography directions. OpenAI has also recruited well-known cryptography experts, such as Harvard professor Boaz Barak, who is deeply involved in the issue of information confidentiality boundaries in cryptography and is now a member of the OpenAI technical team.
Green predicts that cutting-edge AI may achieve a fundamental breakthrough in the elliptic curve cryptography system in the next two years, and elliptic curve is one of the current mainstream encryption algorithms. "Once it comes true, the consequences will be very serious. You don't need to break a lot of algorithms, as long as two of them are broken, a large number of encryption systems will be affected." Green said. A 2025 report by cloud service provider Cloudflare mentioned that the X25519 elliptic curve encryption algorithm is used by the vast majority of the world's Internet traffic.
Green said that if the elliptic curve cryptography system fails, humans will be forced to turn to encryption schemes that are not fully researched and may be less secure.
However, some scholars are skeptical about the prospect of "AI breaking mainstream encryption algorithms."
Ewin Tang, a professor of computer science at Princeton University, said: "These encryption algorithms were designed to be difficult to crack." When she was 18 years old, she achieved remarkable results in the mathematics community and found a classical algorithm that could match the speed of quantum algorithms, overturning the view of quantum theory that had been circulating for decades.
Yuwen Tang believes that
implementation loopholes after the algorithm is written into code
, are often easier to break than the algorithm itself. However, she also mentioned that although well-researched mainstream algorithms are likely to remain secure, some less-researched encryption algorithms, including many quantum-resistant cryptographic algorithms, may be cracked or weakened by AI.In July this year, Anthropic announced its research results: its AI almost broke a quantum-resistant cryptographic candidate algorithm submitted to the National Institute of Standards and Technology (NIST). Youwen Tang said that there will be more similar incidents in the future.
Anthropic spokesperson told "The Information" that the organization's research results in July have been synchronized to the U.S. government and industry partners in advance, and will continue to disclose cryptography-related research to the outside world. The spokesperson wrote: "Once we make significant discoveries, we will disclose them responsibly and take corresponding disposal measures."
Comments