OpenAI’s GPT-6 Cyber: When Your AI Hacking Tool Needs to Know When to Stop
Reading Time: 6 minutesOpenAI’s GPT-6 Cyber scores 48.5% on the ExploitGym benchmark — roughly 51% better than its predecessor — and stays within authorised boundaries 99.93% of the time in lab tests. But a regression in stop-command compliance and gaps in reporting completeness mean the tool still demands careful human oversight before widespread deployment.