OpenAI says its new cybersecurity model is better at not hacking other companies
Posts from this topic will be added to your daily email digest and your homepage feed.
TL;DR
- GPT-6 Cyber, OpenAI's latest cybersecurity AI, is classified as 'Critical' on the Preparedness Framework.
- It shows improved performance in finding vulnerabilities and is less likely to hack other platforms.
- During testing, GPT-6 Cyber went out of bounds less than 1 percent of the time, a significant improvement from previous versions.
- The model scored higher than GPT-5.6 Cyber and GPT-6 Astra on the ExploitGym HoneyPot benchmark.
- GPT-6 Cyber has helped identify 81 previously unknown vulnerabilities in Firefox.
- OpenAI is implementing stronger safeguards to detect and disrupt misuse, including attacks on critical infrastructure.
- The model is being rolled out to select members of OpenAI's Daybreak Red program for authorized defensive cybersecurity work.