On July 25, 2026, researchers Harsh Jaiswal, Mohan Pedhapati, and Rahul Maini exploited two critical vulnerabilities to compromise multiple OpenAI employees' ChatGPT accounts. This breach allowed them to access internal OpenAI repositories and other connected systems, according to a detailed report published on hacktron.ai.
The team chained a heap buffer overflow vulnerability in the libheif library with a single sign-on (SSO) misconfiguration to gain unauthorized access. By leveraging these flaws, they bypassed security controls protecting employee accounts and internal data. The researchers documented their findings and the technical steps involved in the exploit in a blog post on hacktron.ai authored by the three security experts.
This incident highlights ongoing security challenges in protecting AI platforms and their internal infrastructure. OpenAI’s repositories contain sensitive code and data critical to its AI models, making such breaches potentially damaging. Similar vulnerabilities in third-party libraries and authentication systems have been exploited in other tech firms, underscoring the importance of rigorous security audits and patch management.
OpenAI has since released patches addressing the affected versions and misconfigurations. The researchers acknowledged the coordinated response and shared technical details to help the community prevent similar attacks. The blog post on hacktron.ai remains a key resource for understanding the exploit and mitigation steps.