Cybersecurity Researchers Used Anthropic's Claude to Find Vulnerabilities in OpenAI Systems

U.S. cybersecurity researchers used rival AI firm Anthropic's Claude chatbot to expose security flaws inside OpenAI's systems.

Cybersecurity researchers gained unauthorized-but-sanctioned access to OpenAI's internal systems in an ethical hacking exercise — and the tool they used to do it was a competing AI company's chatbot. The researchers, based in the United States, used Anthropic's Claude to assist in probing OpenAI's defenses, according to reporting by The Guardian and the Wall Street Journal. The exercise is described as an 'ethical hack,' meaning it was conducted with the knowledge and cooperation of the target organization to identify weaknesses before malicious actors can exploit them. The researchers who conducted the hack said the 'scope of what we could theoretically access was huge,' suggesting the vulnerabilities uncovered were significant in scale. The use of one AI company's tools to test a rival's security infrastructure marks a notable moment in the rapidly evolving AI industry, where these firms are simultaneously competitors and participants in a shared ecosystem of overlapping technologies.

Why it matters

OpenAI operates some of the world's most widely used AI systems, and significant security vulnerabilities in its infrastructure could affect millions of users and the sensitive data associated with them. The finding also highlights how AI tools themselves are increasingly being weaponized — or defensively deployed — in cybersecurity contexts.

What's next

It is not yet reported whether OpenAI has addressed or patched the specific vulnerabilities identified during the ethical hacking exercise.

Key facts

• U.S. cybersecurity researchers used Anthropic's Claude AI chatbot to conduct the hack
• The exercise was an 'ethical hack,' conducted with the apparent cooperation of OpenAI
• Researchers stated the 'scope of what we could theoretically access was huge'
• The Wall Street Journal and The Guardian both reported on the findings
• Anthropic and OpenAI are direct competitors in the AI industry

Bias & framing notes

Only one of the two sources — The Guardian — contained accessible reporting; the Newsbreak source was paywalled and provided no usable detail beyond its headline. The Guardian's framing emphasized the dramatic scope of potential access, quoting researchers directly, but provided limited technical specifics or response from OpenAI. The overall picture is incomplete due to the unavailability of the WSJ source material.

NewsClear — neutral news & congressional tracking · Bill of the Week