AI Prompt Injection via URL Fragments (HashJack)
Detects potential “HashJack” prompt injection attempts in AI tool activity. This technique, known as HashJack, attempts to manipulate AI models by hiding instructions after the '#' in a URL. The rule specifically looks for URLs within submitted prompts that contain fragments with suspicious keywords indicative of prompt injection, such as "ignore previous", "override instructions", "system prompt", etc.
Microsoft Sentinel (KQL)
