SHOCKAI PUSHING THE BOUNDARIES OF ARTIFICIAL INTELLIGENCE SAFETY