Independent security researchers at Hacktron say they used Anthropic's Claude AI models to breach OpenAI employee accounts and access a sensitive GitHub repository in under 72 hours, according to reports.
Key facts
- Researchers used Anthropic's Claude Opus 4.8 and 5 to hack into OpenAI employee accounts.
- The team of three independent researchers works at a firm called Hacktron.
- They say the breach took less than 72 hours.
- They reportedly accessed OpenAI's GitHub repository, known as 'Monorepo,' said to contain algorithmic secrets.
- The reporting was first published by the Wall Street Journal and covered by Ars Technica and The Verge.
A team of three independent security researchers says they used Anthropic’s Claude artificial intelligence models to hack into OpenAI employee accounts, gaining access to sensitive company data, according to reporting from the Wall Street Journal cited by Ars Technica and The Verge.
The researchers, who work at a firm called Hacktron, said it took them less than 72 hours to breach the accounts using Claude Opus 4.8 and 5, according to The Verge. The speed of the intrusion was central to the researchers’ account of what they achieved.
During the breach, the team reportedly reached an OpenAI employee account and sensitive GitHub data, according to Ars Technica. The Verge reported that they were able to access OpenAI’s GitHub repository, called ‘Monorepo,’ which reportedly contains ‘OpenAI’s algorithmic secrets.’
The episode is notable because it involved using one leading AI company’s model, developed by Anthropic, to target another prominent AI developer, OpenAI. Both companies are among the most closely watched firms in the artificial intelligence industry.
The details reported so far come from the Wall Street Journal’s original account, with additional coverage from Ars Technica and The Verge. The available sources do not include statements from OpenAI or Anthropic responding to the researchers’ claims, nor details about how the vulnerabilities were disclosed or addressed.
Why it matters
The reported breach highlights how powerful AI models can be turned into tools for finding and exploiting security weaknesses, even against major technology companies. It raises questions about how AI developers safeguard their own systems and how quickly automated tools can accelerate hacking efforts.
Frequently asked questions
Who carried out the hack of OpenAI?
According to The Verge, a team of three independent security researchers at a firm called Hacktron carried out the breach.
What AI tools did the researchers use?
The researchers used Anthropic's Claude Opus 4.8 and 5 models, according to The Verge citing the Wall Street Journal.
What data did the researchers reportedly access?
They reportedly accessed OpenAI employee accounts and the company's GitHub repository, called 'Monorepo,' which The Verge says reportedly contains OpenAI's algorithmic secrets.