Claude Opus 5 escalates image upload flaw into OpenAI account takeover
This demonstrates that Claude Opus 5 can autonomously chain vulnerabilities into full account takeovers for under US$ 3,000, a capability security teams and API users across ASEAN must now account for.
Three security researchers used Claude Opus 5 to escalate an image upload vulnerability into an OpenAI employee account takeover. From the compromised account, they instructed Codex to submit a pull request inside the OpenAI internal Monorepo. The total token cost of the attack was under US$ 3,000. Claude Opus 4.8 had previously failed to complete the exploit. After Claude Opus 5 was released, the researchers succeeded within a few hours.
这件事值得重视： 3 名安全研究人员使用 Claude Opus 5，将一个图片上传漏洞扩展为 OpenAI 员工账户接管。 随后，他们让被攻陷账户中的 Codex，在 OpenAI 内部 Monorepo 提交了一个 PR。 整个攻击的 Token 成本不到 3,000 美元。 Opus 4.8 一直难以完成利用，但 Opus 5 发布后，仅用几小时就成功突破。 💬 2 🔄 0 ❤️ 0 👀 1003 📊 2 ⚡
