Security researchers in OpenAI's bug bounty program hacked OpenAI in July and accessed its "monorepo" on GitHub, using Opus 4.8 for cybersecurity and Opus 5

If you pay for a coding assistant, its underlying dependencies are now more likely to be vetted for security flaws.

Security researchers from Hacktron AI successfully compromised OpenAI's internal GitHub repositories on July 25, 2026, by chaining two critical vulnerabilities. The exploit chain began with a heap buffer overflow in the libheif library, which was used by Discourse, the forum platform hosting community.openai.com. This vulnerability allowed for remote code execution on the forum. Researchers then exploited a Single Sign-On (SSO) misconfiguration in OpenAI's identity infrastructure, which allowed them to take over ChatGPT and Codex accounts of OpenAI employees through the compromised forum. By accessing an employee's Codex account, which was connected to OpenAI's GitHub organization, the researchers were able to open a proof-of-concept pull request in OpenAI's internal monorepo, openai/openai, to demonstrate their access without viewing sensitive data. The entire process, from initial discovery to accessing the repositories, took less than 72 hours. OpenAI was notified immediately and issued a $6,500 bounty for the OpenAI-side finding.

The successful hacking of OpenAI's monorepo highlights a critical supply chain risk: the widespread reliance on common, often unvetted, open-source libraries like libheif. The research indicates that vulnerabilities in seemingly minor components, especially those handling user-uploaded content, can cascade into severe security breaches affecting major AI companies. This incident underscores the challenge of securing complex software ecosystems where a single overlooked library can become a gateway for attackers, potentially impacting various services connected through SSO and internal developer tools.

The rapid development and deployment of exploits, aided by advanced AI models like Claude Opus 5, signal an accelerating arms race in cybersecurity. Hackers can now leverage AI to find and weaponize vulnerabilities at unprecedented speeds, forcing organizations to re-evaluate their defensive strategies and patch management timelines. The incident also raises questions about the security implications of AI models themselves being used in exploit development, suggesting that future security research and defense will increasingly involve AI-powered tools and adversarial AI techniques.