Researchers Hack OpenAI Accounts Using Anthropic's Claude AI
A US cybersecurity startup breached OpenAI employee ChatGPT accounts with Claude's help, exposing potentially vast internal access.
Cybersecurity researchers at a US-based startup have successfully penetrated OpenAI's defenses by enlisting a rival artificial intelligence tool — Anthropic's Claude chatbot — to carry out the operation, raising fresh concerns about security vulnerabilities at one of the world's most prominent AI companies.
The team compromised multiple OpenAI employee ChatGPT accounts, using that initial foothold to work toward accessing software caches belonging to their targets. Researchers involved in the effort described the potential exposure in stark terms, noting that the scope of what they could theoretically have reached was substantial.
Read more Warren Buffett, 96, Steps Down as Berkshire Hathaway Chair →
The incident represents the latest in a series of security-related episodes to draw scrutiny toward OpenAI, a company that sits at the center of the global AI industry and handles sensitive data across millions of users and enterprise clients. That a competing company's AI system served as an instrument in the breach underscores the double-edged nature of increasingly capable AI tools.
The findings highlight how advanced language models can be leveraged not only for productivity but also as vectors in cybersecurity exploits — a tension that researchers and policymakers have flagged as AI capabilities accelerate. The episode adds urgency to broader industry debates about safeguarding AI infrastructure against both external attackers and AI-assisted threats.
Continue reading at Business | The Guardian.
