Researchers hack OpenAI's help forum using Anthropic's Claude AI in authorized test

Researchers from Hacktron AI, a start-up, hacked into OpenAI's systems during an authorized security exercise. They exploited a vulnerability in OpenAI's public help forum, which was fixed within 14 hours. OpenAI paid them a $6,500 bounty. The researchers then used Anthropic's Claude Opus 5 AI model to develop a functional attack in just three hours. This incident highlights the growing concern that AI tools are making sophisticated cyberattacks faster and cheaper to execute. Hacktron plans to conduct similar tests on other companies.