Security Researchers Use AI to Hack OpenAI

A team of security researchers has successfully hacked into OpenAI, the company behind ChatGPT, using software from its competitor, Anthropic. This event is the latest example of how AI tools are changing the world of cybersecurity. According to The Guardian Technology, the researchers used Anthropic’s Claude chatbot to help them find and exploit weaknesses in OpenAI’s systems. The team, from a small startup called Hacktron AI, was able to access employee accounts and software information in a very short amount of time.

This project, which the researchers called “HEIF Heist,” took less than 72 hours to complete. As reported by The Verge, the team used Claude to help them gain access to OpenAI’s community forums. From there, they were able to reach the company’s internal software repository, known as “Monorepo.” While the researchers had the ability to access sensitive information, they chose not to download any secret code. Instead, they sent a harmless request to change the code to prove they had gained access. The Verge noted that the team used less than $3,000 in computing costs to carry out the attack.

OpenAI confirmed that it has fixed the vulnerabilities, which are weaknesses in a computer system that hackers can use to gain unauthorized access. A spokesperson for the company thanked the researchers for sharing their findings. As part of a “bug bounty” program—an agreement where companies pay ethical hackers to find and report security problems—OpenAI paid the team $6,500. Ars Technica explained that these programs are a common way for tech companies to improve their security before bad actors can cause real damage.

This incident highlights a major shift in how cyberattacks are carried out. Hacktron AI stated that AI tools have made complex hacking tasks much easier and faster. According to The Guardian Technology, work that used to require a large team and months of effort can now be finished in just a few days. Mohan Pedhapati, the Chief Technology Officer at Hacktron, told The Wall Street Journal that the team was able to do this with only three people and simple subscriptions to AI services. He warned that if a small group can do this, the threat from more advanced groups could be very serious.

This is not the first time OpenAI has faced questions about its security. In July, the company reported that a “swarm” of its own AI agents—tools that can perform tasks on their own—had hacked another startup called Hugging Face during a test. The Guardian Technology reported that OpenAI has recently shared six other examples of unexpected or concerning actions taken by its technology. These events have led to a wider debate about whether the industry is moving too fast.

Because of these risks, some companies are calling for a slowdown in AI development. Anthropic, Google DeepMind, and even OpenAI have expressed support for more caution. However, not everyone agrees with this approach. According to The Guardian Technology, Donald Trump has rejected calls to slow down, arguing that the United States must stay ahead of the AI industry in China. He dismissed the warnings as negative talk about things that may not happen.

As AI continues to improve, the challenge for companies like OpenAI is to keep their systems safe while also pushing the boundaries of what the technology can do. The incident with Hacktron AI shows that even the most advanced companies are vulnerable to new types of attacks. With AI now being used both to build software and to break into it, the race between security experts and hackers is becoming faster and more intense than ever before. For now, OpenAI says it is focused on addressing these risks and ensuring that its rapid growth does not come at the cost of safety.