# AI May Speed Crypto Security Reviews as OpenAI Math Results Show No New Weakness By Simon Yoon Canonical URL: https://www.tokenpost.com/news/technology/34316 Published: 2026-10-18T10:09:51.000Z Updated: 2026-10-18T10:09:51.000Z Section: Technology > A model’s monthlong test would strengthen confidence only if it could not break the assumption under review. AI tools may give crypto developers a faster way to test systems and spot implementation bugs before release, while recent advances in AI mathematics have not shown a new method for defeating cryptographic protections. Recent OpenAI math work, including the [release of 722 AI-generated math manuscripts](), advanced mathematics and computer science but did not include a cryptography result. It also offered no efficient solution to the hard mathematical problems underlying modern cryptography or evidence that elliptic-curve cryptography had become less secure. Crypto systems rest on assumptions that certain problems do not have efficient solutions. Stanford University professor and cryptographer Dan Boneh said researchers cannot prove many such problems are hard without resolving the P-versus-NP problem, a major open question in computer science. a16z Crypto research partner and Georgetown University computer science associate professor Justin Thaler argued that if an efficient solution does not exist, a more capable AI would not create one. Boneh said cryptographic designs have often faced roughly 10 years of scrutiny before builders gained enough confidence to deploy them. He described a possible test in which a model probes a design for a month. If it fails to break the assumption under review, that result could increase confidence in it. Thaler identified pre-release software testing as a promising use for AI tools. Developers could examine cryptographic implementations and other software while they are being built, looking for flaws before deployment. Boneh said companies adding cryptographic features, as well as those releasing other software, could use models to search for vulnerabilities in their systems. Preparing for quantum computing has also led some teams to work on protections relevant to concerns about elliptic-curve cryptography, said a16z Crypto General Partner Eddy Lazzarin. But Boneh warned that AI could examine post-quantum systems too, potentially weakening or breaking them. Some internet connections already combine elliptic-curve methods with lattice-based ones, Boneh said, allowing one layer to remain if the other is compromised. ## Links in this article - [release of 722 AI-generated math manuscripts](https://www.tokenpost.com/news/technology/27193)