OpenAI 内部模型产出 722 篇数学手稿，密码学家担心公钥加密被攻破
OpenAI 内部模型写了 722 篇数学手稿，密码学家都开始算 AI 攻破公钥加密的概率了。
据 The Information 报道，OpenAI 的内部模型已产出 722 篇数学手稿。约翰斯·霍普金斯大学教授 Matthew Green 估计，AI 动摇现有公钥加密可信度的概率为 15%。互联网加密建立在无人能从公钥反推私钥这一从未被证明的假设上，从支付到 WhatsApp 都依赖少数几条椭圆曲线算法。仅 X25519 一项就保护了大部分加密连接，其中 2 条算法被攻破就会波及大量加密流量，而备选方案研究更少、可能更弱。
OpenAI’s internal model produced 722 math manuscripts, and leading cryptographers now worry public key encryption could be next.
~ The information published this piece.
Internet encryption rests on a bet that nobody can work backward from a public key to a private key, and that bet was never proven.
Matthew Green, a professor of computer science at Johns Hopkins gives 15% odds that AI erodes confidence in today’s public key encryption.
Most of the internet, from payments to WhatsApp, leans on a handful of elliptic curve algorithms, and X25519 alone secures most connections. A break in just 2 of them would hit a huge share of encrypted traffic, and the fallback options are less studied and possibly weaker.
