An OpenAI breach on July 25 saw Hacktron AI’s research team, using Anthropic’s Claude, exploit a security flaw to gain access to a core OpenAI software code repository.
During the incident the researchers accessed multiple ChatGPT accounts belonging to OpenAI employees, which allowed them to read and propose changes to code stored on GitHub.
OpenAI has acknowledged the breach, reported that it addressed the vulnerability and awarded a $6,500 bounty to the research team for their findings. An OpenAI spokesperson expressed gratitude for the timely reporting of the issue.
OpenAI breach highlights AI-company vulnerabilities
The event follows a prior incident involving Hugging Face and underscores the susceptibility of AI firms to AI-driven cyberattacks. The episode raises questions about cybersecurity measures in place at AI companies as they face increasingly sophisticated threats.
