Security Researchers Exploit Claude AI to Breach OpenAI Employee ChatGPT Accounts
In a striking demonstration of AI-assisted cyberattacks, security researchers at Hacktron AI successfully accessed an OpenAI employee’s ChatGPT account by leveraging Anthropic’s AI Claude. According to a Wall Street Journal report, the team chained together two critical vulnerabilities to infiltrate OpenAI’s internal systems. The researchers—Harsh Jaiswal, Mohan Pedhapati, and Rahul Maini—detailed their findings in a writeup, revealing that the attack could have allowed them to take over multiple ChatGPT and Codex accounts. The entry point was OpenAI’s community forum, community.openai.com, which they exploited to gain initial access.
The attack began with a vulnerability in an image-processing library, libheif, which the researchers used Claude to build a working exploit for. This flaw, tied to Meta, had a confirmed code-execution exploit according to its maintainers. The team then chained this with a flaw in OpenAI’s sign-in system, specifically a Vercel-related JavaScript flaw that was confirmed in Vercel’s own advisory. By combining these two vulnerabilities, they managed to bypass authentication and take over employee accounts. This gave them entry into OpenAI’s software and ultimately access to internal code repositories, highlighting the severe risks of interconnected vulnerabilities.
The researchers discovered the issue while participating in OpenAI’s bug bounty program and reported it immediately, demonstrating responsible disclosure. The use of Claude to craft the exploit underscores the growing role of AI in cybersecurity—both as a tool for defense and offense. The incident raises concerns about the security of AI platforms and the potential for AI-assisted attacks to escalate. OpenAI has not yet publicly commented on the specific findings, but the bug bounty program’s effectiveness in catching such flaws is evident.
This breach serves as a wake-up call for tech companies to rigorously audit third-party dependencies and authentication mechanisms. The chaining of seemingly unrelated vulnerabilities—from an image library to a sign-in system—shows how attackers can piece together weaknesses to achieve significant access. As AI systems become more integrated into critical infrastructure, the need for robust security practices and proactive bug bounty programs is paramount. The Hacktron team’s work highlights the importance of ethical hacking in uncovering and mitigating these risks before malicious actors can exploit them.
Shimanta Das 
