OpenAI Accounts with Claude: 72 Hour Security Apparatus

While the power of artificial intelligence models increases every day, these capabilities can also be used by malicious actors. One of the striking examples of the last era, one of the three independent security researchers at Hacktron revealed after the access of OpenAI employees’ accounts using Anthropic’s Claude Opus 4.8 and 5 models.

The research team has reached impressive results in very few time:

-**72 hours OpenAI employees managed to overcome authentication systems

• They aimed at the GitHub store called OpenAI “Monorepo”, which contains the algorithm secrets of this warehouse company
• Instead of direct access to internal codes, they sent pull request from an employee account
• Despite the fact that the attack is successfully performed, the researchers stopped at the stage of access to data

According to the report of The Wall Street Journal, this event reveals important questions about the potential security risks of AI models. Advanced language models such as Claude become effective tools in exploiting social engineering and system failures.

This Hadise draws gold that companies should consider AI tools in security testing and attack scenarios more seriously. The researchers have reported the statements they have identified in a responsible way, and they help OpenAI evaluates security posture.