This paper explores how competitive pressure in AI development influences safety choices. Through a behavioral experiment and an evolutionary model, it demonstrates that unsafe development is primarily driven by fear of falling behind and strategic responses to competitors, rather than individual risk preferences alone.