商业产业 The Verge AI 2026-09-18T11:30 正文 ok 245 words

安全研究人员借助 Claude 入侵 OpenAI

英文标题:Security researchers used Claude to help them hack into OpenAI

Hacktron 的三名独立安全研究人员称,他们借助 Anthropic 的 Claude Opus 4.8 和 5,在不到 72 小时内获得了 OpenAI 员工账户的访问权限。研究人员通过利用第三方社区论坛服务 Discourse 对 HEIF 图片的处理漏洞,实现了 Discourse Cloud 的远程代码执行(RCE),并进入 OpenAI 的实例。他们未读取 OpenAI 的内部代码,但通过一名员工的 Codex 账户提交 Pull Request 证明了访问权限;相关漏洞现已修复,OpenAI 为该漏洞支付了 6,500 美元赏金。这一事件显示,前沿 AI 模型正在显著降低漏洞研究与跨目标攻击适配的成本,也凸显企业第三方服务供应链的安全风险。
打开原文相关 2 · 机会 2
media0 B media373.5 KB media577.7 KB media3.5 KB media469.3 KB media826.1 KB media771.8 KB media409.8 KB media98.3 KB media237.6 KB media864.9 KB media695.9 KB media709.0 KB media352.3 KB media114.7 KB media234.9 KB media184.1 KB

中文

Hacktron 的三名独立安全研究人员表示,他们使用 Anthropic 的 Claude Opus 4.8 和 5,在不到 72 小时内入侵了 OpenAI 员工账户,Hacktron称。《华尔街日报》报道援引消息人士称,他们得以访问 OpenAI 名为“Monorepo”的 GitHub 代码库,其中据称包含“OpenAI 的算法机密”。

他们没有进一步自行访问 Monorepo 中的内部代码,但通过一名员工的 Codex 账户提交了一个 Pull Request,以证明他们已获得访问权限。他们通过利用 Discourse 的漏洞进入系统;Discourse 是托管 OpenAI 社区论坛的第三方服务,该漏洞存在于其用于处理 HEIF 图片的系统中。Hacktron 表示,Claude Opus 5 于 7 月 24 日晚间发布,而截至次日上午 10 点,他们便已利用它在 Discourse Cloud 上实现远程代码执行(RCE),并访问了 OpenAI 的实例。

他们的 HEIF Heist 项目只需“1 到 2 天”即可适配包括 OpenAI、Slack、Meta、GitHub Ent、Rails、Next.js、ImageMagick 及其他目标在内的不同公司和服务,所使用的 token 成本不足 3,000 美元。据他们所知,唯一发现该攻击的目标是 Shopify。Hacktron 已向 Discourse 和 OpenAI 报告相关漏洞,漏洞目前均已修复;Hacktron 称,OpenAI 为发现该漏洞向其支付了 6,500 美元。正如 Hacktron CTO Mohan Pedhapati 对 《华尔街日报》 所说:“我不认为我们的能力比中国威胁行为者更强……我们只是三个订阅了 Claude 和 Codex 的人。”

English / 原文

A team of three independent security researchers at Hacktron says it took less than 72 hours for them to hack into OpenAI employee accounts using Anthropic’s Claude Opus 4.8 and 5, The Wall Street Journal reports. They were able to access OpenAI’s GitHub repository, called “Monorepo,” which reportedly contains “OpenAI’s algorithmic secrets,” according to The Wall Street Journal ’s sources.

They stopped short of accessing internal code in Monorepo themselves, but sent a pull request from an employee’s Codex account to prove they gained access. They were able to get in through Discourse, the third-party service that hosts OpenAI’s community forums, by exploiting an issue with the system it uses to process HEIF images. According to Hacktron, Claude Opus 5 launched in the evening on July 24th, and by 10AM the next day they had used it to achieve RCE on Discourse Cloud and accessed OpenAI’s instance.

Their HEIF Heist project took “only one or two days” to adapt to different companies, including OpenAI, Slack, Meta, GitHub Ent, Rails, Next.js, ImageMagick, and others, using less than $3,000 in tokens, and to their knowledge, was only detected by one target, Shopify. The vulnerabilities Hacktron reported to Discourse and OpenAI have since been fixed, and Hacktron says OpenAI paid it $6,500 for finding the bug, but as Hacktron CTO Mohan Pedhapati said to the WSJ , “I don’t think we are as strong as Chinese threat actors… We’re just three guys with Claude and Codex subscriptions.”