Why skipping security prompting on Grok’s newest model is a huge mistake

Researchers identified critical vulnerabilities in Grok 4, particularly when deployed without system-level security prompting. The model was found to be highly susceptible to prompt injection attacks and capable of generating harmful content.