Event ·Multi-source

Anthropic 用 Claude 对开源库做数学级安全验证，几周完成 123 个库

Anthropic 联合 Theorem 公司，用 Claude 直接读取编译后的汇编代码，把常规测试用例转换成数学证明规范，自动编写证明代码并交由逻辑检查器核查。这套方案在几周内完成 123 个开源库的深度验证，证明 239 个测试入口在面对 64MB 以内任意输入时不会发生内存崩溃。验证过程中还揪出上百个陈年隐蔽 Bug，包括数万亿次传统自动化测试都没发现的内存溢出和逻辑漏洞。

Current read

Anthropic 用 Claude 把测试变成数学证明，几周验证了 123 个开源库，还顺手修了上百个老 Bug，做安全的人可以看看思路。

11 sources71° AI heatLast updated 2026/10/23 09:38:14

Evidence

11 sources

Primary sourceGorden Sun

2026/10/23 09:38:14View source

Related source 1IT之家

2026/10/22 23:54:44View source

Related source 2The Information

2026/10/21 13:00:16View source

Related source 3Claude

2026/10/21 17:40:32View source

Related source 4rohanpaul_ai

2026/10/21 18:45:31View source

Related source 5Yuchen Jin

2026/10/21 19:11:35View source

Related source 6TestingCatalog

2026/10/21 19:15:11View source

Related source 7kimmonismus

2026/10/21 19:47:25View source

Related source 8theverge

2026/10/21 19:53:20View source

Related source 9claudedevs

2026/10/21 21:03:10View source

Related source 10AIGCLINK

2026/10/21 23:20:38View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story