Researchers Hack OpenAI Staff Accounts Using Anthropic's Claude
A US cybersecurity startup used Anthropic's Claude chatbot to compromise OpenAI employee ChatGPT accounts, exposing potential access to sensitive software caches.
Cybersecurity researchers at a US-based startup have successfully breached OpenAI by leveraging a rival AI tool — Anthropic's Claude chatbot — to compromise employee ChatGPT accounts, according to a report by The Guardian. The operation represents one of the more pointed security incidents to surface at the artificial intelligence company in recent months.
The team gained access to OpenAI staff accounts and initiated a sequence of intrusions that ultimately allowed them to reach their targets' software caches. Researchers involved in the operation cautioned that the potential exposure ran far deeper, describing the "scope of what we could theoretically access" as "huge."
Read more Tom Watson Joins Palantir Amid NHS and Police Contract Fights →
The incident underscores a growing and somewhat paradoxical vulnerability in the AI industry: AI tools themselves can be weaponized to probe and penetrate the systems of competing AI companies. Using one company's chatbot to ethically hack another adds a new dimension to debates about the security posture of firms handling vast quantities of sensitive user and developer data.
The hack is being characterized as an "ethical" exercise — meaning it was conducted with a defined scope and presumably without malicious intent — yet the findings point to systemic risks that extend beyond a single breach. For OpenAI, which has faced a series of scrutiny over internal security practices, the episode adds fresh pressure to address how employee-facing tools and credentials are protected against AI-assisted attacks.
Continue reading at Business | The Guardian.