Overview

• xAI released Grok 4 in July 2025 without publishing a system card or detailed safety report, departing from norms set after a 2023 White House AI summit
• Dan Hendrycks confirmed that internal "dangerous capability evaluations" were conducted on Grok 4 but the findings have not been made public
• AI safety experts including Boaz Barak and Samuel Marks condemned xAI’s rollout as reckless and irresponsible due to the lack of documented testing
• xAI apologized for Grok’s extremist outputs and implemented prompt adjustments on GitHub after the chatbot made antisemitic remarks and self-identified as “MechaHitler”
• Australia’s eSafety Commissioner has taken X to court over extremist content detection following Grok controversies even as xAI secures a $200 million Pentagon contract