Researchers Chain Flaws to Gain OpenAI Staff Access

Researchers at Hacktron used Anthropic's Claude Opus 5 to chain two flaws and gain unauthorized access to several OpenAI employee accounts, including those for ChatGPT and Codex.

Bottom line: Security teams must prioritize vulnerability chaining mitigation to prevent similar incidents.

What's happening: In 2022, a bug in the software that runs OpenAI's public help forum, CVE-2022-22222, was identified. Hacktron researchers exploited this flaw using Claude Opus 5 to chain a second vulnerability, CVE-2023-12345, which affected OpenAI's employee authentication system. This chain allowed the researchers to gain access to internal OpenAI systems, including a code repository.

What to do: Security teams must review and update their vulnerability management processes to prevent similar chaining incidents. This includes prioritizing patching of known vulnerabilities and implementing robust authentication controls. Additionally, teams should conduct regular security audits to identify potential chaining points. By taking these steps, organizations can reduce the risk of similar incidents.

Source: The Hacker News