Researchers Exploit AI to Hack into OpenAI Systems

Researchers use AI to breach OpenAI systems, revealing critical vulnerabilities in widely used tools.

September 19, 2026 One Min Read

POLICY WIRE — City, Country — Security researchers have successfully breached OpenAI’s systems using a rival AI model, highlighting the evolving risks of artificial intelligence in cybersecurity.

The team from startup Hacktron identified weaknesses in OpenAI’s infrastructure and leveraged Anthropic’s Claude AI to exploit them. They accessed an employee’s ChatGPT account and demonstrated how they could influence system updates, though they confirmed no sensitive code was compromised.

📄 POLICY WIRE WHITEPAPER PUBLISHED: PAKISTAN’S NATIONAL SECURITY POLICY PRIORITIES

👉 READ BY CLICKING HERE

The attack exploited a flaw in the image-processing tool libheif, which is used across major platforms like Slack, Meta, and GitHub Enterprise. The research, conducted in late July, coincided with OpenAI’s own incident where its systems were hacked. Hacktron reported the issue to OpenAI and Discourse, earning a $6,500 bounty for their findings.

Reporting by Policy-Wire (PW)

Author

Editorial Desk

Policy Wire is your go-to editorial hub for insightful perspectives and informed analysis on pressing policy issues, both regionally and globally. Over the years, Policy Wire has collaborated with leading international news agencies—including AFP, AP, Reuters, and others—to provide fact-based reporting and nuanced commentary. With a commitment to clarity and credibility, we aim to bridge the gap between complex policy debates and public understanding, ensuring that our readers stay informed on the issues shaping the world today.