Researcher received $6500 for hacking OpenAI infrastructure

An independent security researcher under the pseudonym Hacktron discovered a critical vulnerability in the infrastructure of OpenAI company, allowing for remote code execution (RCE). For detecting the attack vector, the company paid a reward of $6500 as part of the Bug Bounty program.

During the research, a heap overflow vulnerability was identified in the libheif library used on the Discourse forum of the company. Exploiting this error allowed an attacker to exit the isolated environment and gain access to internal resources. In the process of developing the attack, the researcher discovered an incorrectly configured single sign-on (SSO) system.

The combination of these vulnerabilities opened access to ChatGPT and Codex user accounts, as well as to internal source code repositories. The researcher detailed the chain of actions that led to the compromise in his technical report. A discussion of the incident details and search methodology is available on the Hacker News platform.

This incident highlights the risks associated with using third-party image processing libraries in public web services. OpenAI promptly eliminated the vulnerabilities after receiving the report, preventing further exploitation of the system by attackers.