Researchers used Claude to breach OpenAI's internal systems

Researchers applied Claude to breach OpenAI's internal systems. AFP Published September 18, 2026 Updated September 18, 2026 07: 54pm.

Key points

• OpenAI confirmed the flaw was fixed within regarding 14 hours of being notified and paid the researchers a $6, 500 reward.
• The Hacktron researchers remarked they initially employed Anthropic's Claude Opus 4.8 to identify and exploit the software flaw, but struggled to create it work consistently.
• After Anthropic published Claude Opus 5, the researchers stated the newer model produced a working hack within regarding three hours.
• The hackers did not employ Claude Mythos, a more capable Anthropic model that is restricted to a small group of vetted cyber-defence organisations.
• Hacktron remarked the underlying software flaw is not unique to OpenAI and is employed throughout numerous firms' products, including those produced by Slack and Meta.

In practice, a security research firm stated Friday it managed to break into OpenAI's internal systems using the latest software from Anthropic, exposing how rapidly the technology can carry out sophisticated cyberattacks.

Notably, the researchers from security firm Hacktron remarked they identified a security flaw in OpenAI's public support forum, run by the Discourse platform, that allowed them to take control of the site.

"We immediately reported the initial vulnerability to OpenAI and Discourse and worked with them to coordinate the patch, " Hacktron remarked in a blog post.

"We appreciate their attention to detail and fast resolution of this issue, " the post continued.

"We thank the researchers for contacting us and sharing their findings. We narrowed the permissions on Community sign-in tokens and revoked affected tokens and sessions, " remarked Drew Pusateri, an OpenAI spokesperson.

For context, the Hacktron researchers remarked they initially employed Anthropic's Claude Opus 4.8 to identify and exploit the software flaw, but struggled to create it work consistently.

Meanwhile, the hackers did not employ Claude Mythos, a more capable Anthropic model that is restricted to a small group of vetted cyber-defence organisations.

Anthropic has described Mythos as having the strongest cybersecurity capabilities of any model it has built.

Hacktron remarked the underlying software flaw is not unique to OpenAI and is employed throughout numerous firms' products, including those produced by Slack and Meta. The firm remarked it is continuing similar tests at other businesses.

Meanwhile, the case adds to growing concern among security experts that AI tools are making it faster and cheaper to carry out sophisticated cyberattacks that once required specialised teams and months of work.

Taken together, the developments around researchers used Claude to breach OpenAI' s internal systems point to a situation that is still moving, and the coming days should bring more clarity.