【#Tech24H】OpenAI recently posted a blog to officially introduce GPT-6 Cyber, a dedicated AI model built for cybersecurity. The model delivers marked improvements over its predecessor in discovering, verifying and remediating software vulnerabilities, offering a new technical tool for cybersecurity research. According to the company, GPT-6 Cyber scored 48.5% on the ExploitGym benchmark, compared with 32.1% for the prior GPT-5.6 Cyber — a 51% relative performance gain. This benchmark mainly evaluates the model’s capability to analyze and exploit remote vulnerabilities. On safety guardrails: without dedicated safety prompts configured, the model refrains from unauthorized access to services outside its authorized scope at a rate of 99.93%. With safety prompts enabled, it achieves a 100% success rate in preventing out-of-bounds exploitation.