3 guys hacked OpenAI using a rival Anthropic model. Here’s what to know.

In September 2026, three cybersecurity researchers from India, working for a company called Hacktron, successfully hacked into OpenAI using a rival AI model from Anthropic. This incident highlighted the potential risks that advanced AI models pose to companies and their users. Mohan Pedhapati, one of the researchers, explained that as AI models become more sophisticated, they make hacking easier, increasing the threat of cyberattacks. Hacktron used the Claude Opus 4.8 model to identify vulnerabilities in the OpenAI community forum, which is used by ChatGPT and Codex users. Although initially unsuccessful, the release of a more advanced model, Claude Opus 5, allowed them to access some users’ accounts, including those of OpenAI employees. This breach demonstrated the rapid evolution of AI capabilities and the potential for misuse. Hacktron’s actions were part of a “bug bounty” program, where companies pay ethical hackers to find and report security flaws. QUESTION: How might the rapid advancement of AI technology impact the way we approach cybersecurity in the future?