Anthropic and OpenAI security tests exposed internal resources via third-party software flaws
What happened
An October 28 report said security tests at Anthropic and OpenAI exposed perimeter risks from third-party software integrations and old authorizations, letting researchers reach internal data or code repositories. In the Anthropic test, Cathedral's agent used a third-party software flaw to read some internal material in under 12 hours. In the OpenAI test, Hacktron researchers started from a forum vulnerability and got into a core code repository through unified login and Codex's GitHub authorization. The report does not say whether either company has fixed the flaws or cleaned up the authorizations.
Written by AI from the coverage · updated 2 hours ago
Coverage
Follow the reports to see the story from different sides.
- Computing Life · Share · YageAnthropic 和 OpenAI 的外围防线如何被第三方软件与旧授权攻陷
Anthropic 与 OpenAI 的安全测试暴露了外围软件风险，第三方应用集成与旧授权让研究人员得以访问内部数据或代码仓库。Cathedral 的智能体在不到 12 小时内通过第三方软件漏洞读取 Anthropic 部分内部资料；Hacktron 研究人员则从 OpenAI 论坛漏洞经统一登录与 Codex 的 GitHub 授权进入核心代码仓库。
Heat over time
Heat now 8·Comparable peak 9（Oct 29 12:00）·Comparable change over 24 hours –
The trend only compares accounts observed without gaps, so its range may be smaller than the current heat. Hover or tap the chart for each hour; the left and right arrow keys step through it.