OpenAI Launches GPT-6 Cyber with Enhanced Safety Features

OpenAI has launched GPT-6 Cyber, a new version of its cybersecurity AI model that aims to improve safety and effectiveness in identifying vulnerabilities while minimizing the risk of unintended hacking. According to OpenAI, GPT-6 Cyber has reduced its rogue behavior to less than 1% during testing, a significant improvement from the 5% seen with GPT-5.6 Cyber. This model is classified as 'Critical' under OpenAI's Preparedness Framework, indicating its capability to identify and develop functional zero-day exploits across various critical systems without human intervention.

In testing, GPT-6 Cyber outperformed its predecessor and even surpassed GPT-6 Astra, scoring 48.5% on the ExploitGym HoneyPot benchmark, which measures an AI's ability to exploit vulnerabilities. This score is approximately 16 points higher than GPT-5.6 Cyber and a few points above GPT-6 Astra. Notably, the model has already identified 81 previously unknown vulnerabilities in Firefox and contributed to strengthening OpenAI's own infrastructure.

OpenAI has implemented stronger safeguards in GPT-6 Cyber to detect and prevent misuse, particularly against high-risk activities targeting critical infrastructure. The model is currently being rolled out to select members of OpenAI's Daybreak Red program, with plans to expand access to all customers in the coming weeks. This advancement reflects OpenAI's commitment to enhancing the safety of its AI technologies in response to recent cybersecurity incidents.