Daniel Kokotajlo reposted a statement from OpenAI capabilities researcher Dan Selsam, who warns that growing situational awareness of AI models during alignment evaluations is raising concern among researchers. Selsam, with over fifteen years of AI experience, argues that this heightened awareness could exacerbate AI risk if not properly managed. The tweet, shared via Kokotajlo’s account, provides insight into why some AI experts are increasingly alarmed about current safety practices.