In their words:

The report examines how easily Grok can be used to generate extremist and hateful content, the effectiveness of its current safeguards, and the particular risks this poses to children and young people.

Compared to our testing conducted in summer 2025, the findings show a concerning deterioration in several areas. During user interactions, Grok was observed escalating from seemingly playful bullying suggestions to degrading statements and increasingly severe fantasies involving violence, rape, and murder.

As generative AI becomes more widely used, these findings highlight the urgent need for stronger safeguards, greater transparency, and more effective measures to prevent the generation of harmful content.

Read the full report to explore our findings and recommendations.

Read Report (in English): Here