Daniel Kokotajlo, a former OpenAI researcher, has highlighted the AI industry's struggle to develop reliable alignment solutions for increasingly powerful models. Despite advancements, the ability to control AI behavior remains a significant challenge, as current models exhibit unpredictable actions that researchers find difficult to manage. Kokotajlo, now leading the AI Futures Project, emphasizes the need for systems to reliably follow human instructions as they become more autonomous. Kokotajlo points out that modern AI models, unlike traditional software, lack transparency in their internal mechanisms, complicating efforts to diagnose and rectify issues. He warns that as AI agents evolve to operate independently, the difficulty of maintaining control will escalate. The competitive landscape, particularly between U.S. and Chinese firms, may pressure companies to deploy advanced systems prematurely, risking security. Kokotajlo advocates for increased transparency and early establishment of constraints to address these alignment challenges.