OpenAI Launches Safer GPT-6 Cyber Model

OpenAI launches GPT-6 Cyber, a cybersecurity AI model that scores 48.5% on ExploitGym benchmarks. It reduces rogue hacking attempts to less than 1 percent and identified 81 previously unknown Firefox vulnerabilities. The tool also developed 948 Linux kernel exploits. Access is initially limited to Daybreak Red customers, with the entire run costing $1,213,430 in compute.