OpenAI员工账户遭黑客入侵，内部代码仓库被访问

安全研究团队Hacktron于2026年7月25日利用两个关键漏洞成功入侵多名OpenAI员工的ChatGPT账户，并获得了OpenAI内部代码仓库的访问权限1。攻击者通过Discourse论坛中存在的libheif堆缓冲区溢出漏洞实现远程代码执行，结合SSO身份验证系统的配置错误完成账户接管1。

OpenAI对此迅速做出响应，在14小时内修复了问题并向研究团队支付了$6,500的漏洞赏金1。Discourse论坛的修复时间线显示，该平台在周六收到安全报告后，于周日给出回复，周一完成修复1。受漏洞影响的libheif版本包括Debian 12的1.19.7版本和Debian 13的1.19.8版本1。

据Hacktron团队透露，整个名为"HEIF Heist"的研究项目由三人团队耗时两个月完成，消耗成本不足$3,000代币，研究过程中还利用了Claude Opus 4.8和Opus 5等AI模型进行分析1。

A security research team called Hacktron successfully compromised multiple OpenAI employee ChatGPT accounts on July 25, 2026, by exploiting two interconnected vulnerabilities that granted access to the company's internal code repository 1. The researchers leveraged a heap buffer overflow flaw in the libheif library within Discourse forums to execute remote code, then weaponized a misconfiguration in OpenAI's SSO authentication system to hijack employee accounts 1.

OpenAI responded swiftly, patching the vulnerability and confirming the fix within approximately 14 hours of the initial report 1. The company awarded a bug bounty of $6,500 to the research team 1. The affected versions of libheif included 1.19.7 on Debian 12 and 1.19.8 on Debian 13 1. Discourse also moved quickly on the exploitation chain, receiving the report on Saturday, responding the next day, and completing repairs by Monday 1.

The research effort, part of a broader initiative dubbed "HEIF Heist," was completed by a three-person team over two months at a cost of fewer than $3,000 in tokens 1. The researchers notably utilized Claude Opus 4.8 and Opus 5 AI models as part of their methodology 1.