US cybersecurity researchers ethically hacked OpenAI systems using Anthropic’s Claude chatbot and OpenAI’s own GPT-5.6 Sol, exposing vulnerabilities in employee accounts and related software infrastructure. The researchers accessed several staff ChatGPT accounts through an internal discussion forum and then reached OpenAI’s GitHub service, highlighting how AI tools can make sophisticated cyberattacks faster and easier.

The team said the potential access was extensive, though it did not download repository code, and reported the weaknesses through OpenAI’s bug-bounty programme, receiving $6,500. OpenAI said it had fixed the exploited vulnerabilities; the incident comes amid other concerns about the company’s AI systems, wider warnings over rapid development, and renewed calls from several leading AI firms for greater caution.