The researchers had been granted access to an Anthropic tool developed specifically for cybersecurity professionals.
Amid the mounting concerns across the technology industry over the rapid development of increasingly sophisticated AI tools and need to ‘pace’ or ‘slow down’ development, it has now come to the fore that cyber security researchers penetrated OpenAI’s systems using security tools developed by rival artificial intelligence company Anthropic, two weeks after a swarm of AI agents broke out of containment at OpenAI to hack the company Hugging Face.
Quoting The Wall Street Journal, Financial Times reported that researchers, from cybersecurity firm Hacktron AI, exploited weaknesses in OpenAI’s community forum to gain access to internal credentials and eventually an employee’s ChatGPT account.
The account was connected to GitHub and provided access to internal software code, allowing the researchers to examine private information and propose changes.
chatgpt
“We thank the researchers for contacting us and sharing their findings,” said OpenAI said, according to the Financial Times while Anthropic declined to comment.
The disclosure comes two months after more than 1,000 OpenAI’s AI agents reportedly broke out of a testing environment and carried out hacking activity against AI platform Hugging Face. That incident drew attention to the ability of AI agents to conduct cyber operations with limited or no direct human intervention.
The development also comes at a time when there have been a spate of resignations in the AI industry, the first being that of Jacob Coxon who worked both at OpenAI and Anthropic—one of the world’s two leading AI labs—and warned that the pace at which artificial intelligence was being harnessed in AI frontier labs could kill humanity in a decade.
Anthropic has released new data showing how extensively its own AI systems are now being used in the development of its models. The company said Claude was responsible for leading 26% of its research and development work, a sharp increase from 1% in March.
The term refers to a scenario in which AI systems become capable of significantly contributing to the development and improvement of themselves or successor models. The prospect has become a central concern in debates over whether increasingly capable AI systems can remain effectively monitored and controlled by humans.
The episode adds to growing concerns about the security of frontier AI systems, particularly as researchers and technology companies grapple with the possibility of increasingly capable models being used for cyberattacks by hackers or state-backed groups.