“Nikesh Arora believes that the safety guardrails on current AI models are not robust enough and can be easily bypassed.”