Topline
A group of cybersecurity researchers managed to breach OpenAI’s internal system using rival Anthropic’s Claude AI software as part of a bug bounty program, highlighting the growing threats posed by advanced AI models amid broader safety warnings about the technology from top industry executives and researchers.
Key Facts
Researchers at the security startup Hacktron AI to an OpenAI employee’s ChatGPT account, which then gave them access to the company’s internal code on GitHub.
The first reported the incident, noting the team was participating in an OpenAI bug-hunting program and that the company paid the researchers a $6,500 bounty for disclosing their work.
In its , Hacktron said they first tried to use Anthropic’s Claude Opus 4.8 to attempt the breach, but it “struggled across several sessions to produce a working exploit.”
However, once Anthropic launched Opus 5, they tried again and succeeded, and noted that “every new model is getting increasingly capable.”
According to the Journal, the researchers had access to a “special version” of Claude which has been made available to “qualified cybersecurity practitioners.”
In a statement to Forbes, OpenAI spokesperson Drew Pusateri said the company thanked the researchers “for contacting us and sharing their findings,” and noted that it had resolved the vulnerability.
Surprising Fact
In their report, the researchers said the OpenAI hack took only a “few days for an agent, and just a few hours of human time,” to pull off. The breach was part of “”, a broader research project into a vulnerability of how many software and services process certain image files. This exploit has been able to find vulnerabilities with Slack, Zoom, Meta and others and only took two months, three researchers, and cost “less than $3,000 in tokens in total.”
Crucial Quote
In their conclusion note, the researchers wrote, “Software has long benefited from a kind of security through complexity…AI is removing that protection by turning more of this scarce expertise into compute. Work that once required a well-resourced team and months of effort can now be compressed into days…Security assumptions must catch up with attacker capabilities.”
further reading
(Wall Street Journal)
Billionaire Anthropic CEO Urges Competitors To Slow Down AI Development (Forbes)