As AI ethics principles increasingly shape regulations, developers face growing responsibilities to mitigate the risks posed by their applications. In this paper, we present a mixed-methods survey study to examine how developers across various roles perceive and address risks in AI systems. Our survey comprises 414 participants from 23 countries, representing a diverse range of roles involved in AI development. The results reveal significant disconnects in risk mitigation practices. While quality assurance teams are the most proactive in testing for demographic risks, developers often prioritize technical solutions and overlook vulnerabilities. Notably, security experts report the lowest levels of engagement with AI governance initiatives. These disparities point to a fragmented approach to risk mitigation across AI development roles. In response, we propose future research directions and educational strategies.