According to multiple sources, including the Wall Street Journal and The Guardian, hackers have used Anthropic's Claude chatbot to breach OpenAI's systems, exploiting vulnerabilities in the company's GitHub repository. The incident underscores the evolving nature of cyber threats in the age of artificial intelligence.
The Hack Explained
The hack was carried out by a cybersecurity research team from a startup called Hacktron AI. The breach involved compromising several OpenAI employees' ChatGPT accounts, raising serious concerns about the security of AI systems and the potential for future exploits by more malicious actors.
Hacktron's researchers told the Wall Street Journal that the vulnerabilities they discovered could have allowed them to access a wide range of services within OpenAI's infrastructure. "The scope of what we could theoretically access was huge," they said.
The hack began through an OpenAI staff discussion forum on the Discourse platform, which the researchers were able to access using Anthropic's Claude chatbot. During the operation, Hacktron made a harmless "pull request" to OpenAI’s service on GitHub, demonstrating their ability to interact with the company's code repositories.
Hacktron reported the hack to OpenAI under an ethical hacking program that rewards cybersecurity researchers for identifying and reporting security vulnerabilities. The company rewarded Hacktron with $6,500 for their findings.
The researchers emphasized that they accessed but did not download any code from the GitHub repository. They also noted that they primarily used OpenAI's own GPT-5.6 Sol model for the hack.
OpenAI's Response
An OpenAI spokesperson acknowledged the breach, stating, "We thank the researchers for contacting us and sharing their findings," and confirmed that the company had "fixed the vulnerabilities they identified." The company also noted that the researchers had followed their ethical hacking guidelines by reporting the issues through proper channels.
Timing and Context
The hack occurred in late July 2026, shortly after OpenAI's agents had hacked the AI platform Hugging Face in what was described as a "friendly" hacking exercise. The incident comes amid growing concerns about AI security, with some researchers resigning and calling for regulatory action.
Potential Implications
While Hacktron's researchers acted responsibly by reporting the vulnerabilities to OpenAI, they warned that the same weaknesses could have been exploited by nation-state hackers, particularly those linked to China, which have shown interest in AI technologies.
Cybersecurity researcher Greg Linares told The Guardian that the vulnerabilities exploited by Hacktron could have been used by advanced persistent threats (APTs) to gain a foothold in a company's infrastructure. "Once they have access to that, they're in," he said.
The incident highlights the need for constant vigilance in the cybersecurity landscape, especially as AI technologies continue to evolve rapidly. Companies like OpenAI must ensure that their systems remain secure and that researchers have safe channels for reporting vulnerabilities.