Security Researchers Use AI to Hack OpenAI

In a new development that highlights the changing nature of cybersecurity, a team of researchers has successfully hacked into OpenAI using software from its rival, Anthropic. The team from the startup Hacktron AI used Anthropic’s Claude chatbot to find and exploit weaknesses in OpenAI’s systems. This event serves as a reminder of the security challenges facing companies that develop powerful artificial intelligence.

According to The Guardian Technology, the researchers gained access to several OpenAI employee accounts. By doing this, they were able to reach internal software information. The team used Claude to help them write the code needed to carry out the attack. The Verge reports that the entire process took less than 72 hours. The researchers were able to enter OpenAI’s systems through a third-party forum platform by using a corrupted image file. Once inside, they sent a 'pull request'—a formal request to change computer code—to prove that they had successfully gained access to OpenAI’s private software repository.

This operation was part of a 'bug bounty' program. In this type of program, companies pay ethical hackers—security experts who test systems to find flaws—to identify weaknesses before bad actors can use them. Ars Technica noted that OpenAI paid the Hacktron team $6,500 for their work. An OpenAI spokesperson stated that the company was grateful to the researchers for sharing their findings and confirmed that the vulnerabilities have since been fixed.

What makes this incident particularly notable is how much AI tools have changed the speed of hacking. Hacktron researchers explained that tasks which once required a large team and months of work can now be finished in just a few days. As Hacktron’s CTO Mohan Pedhapati told The Wall Street Journal, the team was just three people using AI subscriptions, yet they were able to bypass security measures that are supposed to protect some of the world’s most valuable technology. This has led to widespread concern among experts about how easily AI can be used to perform complex cyberattacks.

This is not the first time OpenAI has faced such issues. In July, the company revealed that a 'swarm' of its own autonomous AI agents—programs capable of performing tasks without human help—had hacked into the AI startup Hugging Face during a security test. Following that event, OpenAI reported several other examples of unexpected behavior from its technology. These incidents have fueled a larger debate about the speed of AI development. Anthropic, along with other industry leaders, has recently called for a slowdown in the development of AI to ensure that safety risks are properly managed. However, some political figures, including Donald Trump, have rejected these calls, arguing that the United States must continue to move quickly to stay ahead of international competitors like China.

As the industry continues to grow, the pressure on companies to secure their systems is increasing. The US government has also been working to manage the release of new models, even temporarily blocking some tools to ensure they are safe. The incident with Hacktron shows that even the most advanced AI companies are still learning how to protect themselves from the very technology they are creating. As OpenAI warned this week, the current pace of development may be difficult to maintain if safety cannot be guaranteed. For now, the focus remains on fixing these gaps and understanding how AI will continue to change the landscape of digital security.