The 'M365 Copilot Jailbreak Attempts' detection rule targets malicious activities intended to manipulate the Microsoft 365 Copilot functionality through prompt injections. The rule is structured to search exported eDiscovery logs for specific keywords associated with attempts to bypass or subvert built-in safety protocols of the Copilot AI. The detection logic scores prompts based on their potential to compromise the system, with higher severity scores assigned to explicit impersonation or rule bypass attempts. It employs a systematic approach to log evaluation, allowing security teams to identify and assess risks associated with user actions that may violate organizational policies regarding AI usage. False positives could arise from legitimate discussions on AI ethics and safety, which necessitates careful analysis of flagged events.