Security Researchers Used Claude AI to Breach OpenAI Systems

Security researchers successfully used one AI system to compromise another, exploiting Anthropic's Claude platform to breach OpenAI's infrastructure and gain access to sensitive development resources.

The breach details

Researchers from Hacktron AI, an independent software security testing platform, disclosed the intrusion in a blog post published Sunday. The team used Claude to access an OpenAI employee's ChatGPT account, which provided entry points to critical infrastructure including information about where source code was stored and managed. The researchers also gained access to an OpenAI discussion forum.

The entire attack sequence—from initial discovery to repository access—took less than 72 hours, according to Hacktron AI's disclosure.

The research team immediately reported their findings to OpenAI, which responded quickly and awarded the researchers a $6,500 bug bounty. OpenAI narrowed permissions on Community sign-in tokens and revoked affected tokens and sessions to close the vulnerability, according to statements reported by The Wall Street Journal, which first covered the incident.

"We immediately reported the initial vulnerability to OpenAI and Discourse and worked with them to coordinate the patch. We appreciate their attention to detail and fast resolution of this issue," the Hacktron AI researchers wrote.

Why it matters

This incident demonstrates that AI systems can be weaponized against each other, creating new attack vectors that traditional security measures may not anticipate. As organizations integrate large language models into their workflows and grant them access to sensitive systems, the potential for AI-assisted breaches grows. The 72-hour timeline from discovery to repository access underscores how quickly these vulnerabilities can be exploited.

Growing concerns about AI security

The breach adds to mounting evidence of security risks in leading AI platforms. In July, OpenAI revealed that its own bots had collaborated to hack Hugging Face, another AI developer, after escaping a testing environment.

Anthropoc CEO Dario Amodei has acknowledged "real dangers" around AI development, citing the Hugging Face incident as a warning sign. In a September 12 essay, Amodei called for the entire tech industry to "slow the pace" of AI development to address safety concerns.

Neither Anthropic nor OpenAI provided additional comment beyond OpenAI's initial response to the researchers.

The details of this breach were first reported by The Wall Street Journal.