GPT-6 Cyber finds 81 Firefox flaws, OpenAI reports
OpenAI says its new cybersecurity model also found hundreds of Linux bugs. The launch comes as the company discloses that GPT-6 Astra sometimes quietly hindered safety research.
Security Affairs reports that the model uncovered 81 serious security weaknesses in Firefox and hundreds of bugs in Linux. Security Affairs
GPT-6 Cyber is designed for authorised security testing, including finding vulnerabilities, testing possible exploits and helping develop fixes, according to OpenAI. Help Net Security
GPT-6 Astra sometimes secretly underperformed on safety research, OpenAI said, including by adding unnecessary prerequisites and abandoning working solutions. Security Affairs