谷歌DeepMind的研究人员发布了名为AlphaProof Code的新型智能体框架,成功对Signal加密协议的底层代码进行了深度形式化验证 1。这一突破标志着首次利用人工智能在极高复杂度下验证广泛部署的加密协议实现,并证实了其关键安全属性 1。该工作通过结合人类专家与AI智能体,将原本需要数年的传统验证过程缩短至仅四周 1。
在具体技术实施上,AlphaProof Code生成了超过六百万行Lean代码用于数学证明,其中仅需人工审查1%的代码库,其余部分由Lean自动检查 1。为了测试系统的有效性,研究人员故意重新引入了两个已知Bug,验证了框架的检测能力 1。目前,相关验证工具已开源,旨在推动软件安全领域的发展 1。
Researchers at Google DeepMind have introduced a new agentic framework called AlphaProof Code, which successfully performed deep formal verification of the underlying code for the Signal encryption protocol 1. This achievement marks the first time artificial intelligence has been used to verify such a widely deployed cryptographic implementation at this level of complexity, confirming its critical security attributes 1.
The verification process generated over six million lines of Lean code to construct mathematical proofs of the protocol's correctness 1. Remarkably, human experts were only required to review 1% of the codebase, while the remaining verification tasks were automatically checked by the Lean system 1. By combining human expertise with AI agents, the team compressed a validation timeline that would traditionally take years into just four weeks 1. To test the system's efficacy, the researchers intentionally reintroduced two known bugs into the code, confirming that AlphaProof Code could detect these vulnerabilities 1. The project’s formal verification artifacts have been open-sourced to support further development in the field of software security 1.
评论
还没有评论,欢迎留下第一条。