The model is meant to help security researchers find vulnerabilities, test how they could be exploited and fix flaws before attackers do, OpenAI said