
Hacktron breaches OpenAI using Anthropic's Claude to exploit software flaw
The story in brief
Hacktron researchers used Anthropic's Claude Opus 4.8 to exploit a software flaw in OpenAI's systems, which was fixed within 14 hours and resolved with a $6,500 reward. The flaw, not unique to OpenAI, was also found in products from Slack and Meta. After upgrading to Claude Opus 5, the researchers achieved a working hack in three hours.