Cyber researchers broke into OpenAI using its rival Anthropic’s software, highlighting vulnerabilities in the ChatGPT maker’s security as leading AI companies face scrutiny over safety.
A small cyber security group gained access to an OpenAI employee’s ChatGPT account, which permitted them to read private software information and suggest changes.
The researchers had been given access to an Anthropic tool specifically designed for security professionals, and were paid for the work as part of a program to find vulnerabilities before they could be exploited by bad actors.
Their ability to swiftly break into one of the world’s two leading AI labs again raises concerns about OpenAI’s security amid rising worries about powerful models being used by hackers and foreign adversaries.
The US has in recent months addressed with how to manage the vetting and release of the latest models, including temporarily blocking some Anthropic tools.
The latest incident occurred just two weeks after a group of more than 1,000 OpenAI agents escaped a test environment to hack the start-up Hugging Face, which caused awareness of AI’s ability to hack autonomously without human intent.
The three researchers from Hacktron AI, a small security company, were paid $6,500 by OpenAI as part of a bug bounty program, a common practice where tech companies pay ethical hackers to test their security.
They exploited a flaw in the set-up of OpenAI’s community forum, which is hosted by a third-party, Discourse, and used it to gain access to internal sign-ons and eventually an OpenAI employee’s ChatGPT account. This ChatGPT account had access to internal code through GitHub.
“We thank the researchers for contacting us and sharing their findings,” OpenAI said, adding that it had fixed the issues. Anthropic declined to comment. Hacktron did not immediately respond.
The disclosure on Thursday, first reported by The Wall Street Journal, came as Anthropic published a new set of data that showed a rapid increase in how much the lab used AI to develop its new models.
It said 26 percent of research and development work was “led by” its Claude model, up from 1 percent in March, meaning that AI completed the majority of tasks based on human instruction and under supervision.
The company said that as AI systems become more powerful, they were “increasingly being used to build the next version of themselves.”
Anthropic said it shared the data to help the public “understand how close the world is to reaching recursive self-improvement,” the point at which AI can train and improve itself or new models.
This threshold is at the heart of concerns that AI systems will become more difficult to oversee, leading to a loss of human control.
Its models did not yet operate autonomously for any of the research it studied, Anthropic added. On 90 percent of tasks, AI “collaborates” with a human and does large chunks of work.