A recent study from the University of Manchester highlights that the main challenges in AI ethics may not be about what individual engineers know or care about, but rather about the structures and systems within the organizations they work for. The research, based on in-depth interviews with AI and software engineers across various industries, found that while engineers are often aware of ethical risks—such as biased outputs, unexplainable decisions, and inaccurate results—they frequently lack the authority, support, or incentives to act on these concerns. Many expressed a desire to implement safeguards but felt constrained by the environments in which they work. The researchers refer to this situation as "ethical awareness without ethical agency"—engineers who understand what is right but are unable to make it happen. This issue was presented at the 10th Data for Policy Conference and raises questions about whether current oversight systems measure real ethical behavior or just the appearance of it. The study found that factors such as bureaucratic compliance processes, tight deadlines, and reward systems that favor speed over thoroughness create a situation the paper calls "compliance theater"—where organizations appear ethical on paper but do not follow through in practice. Alessia Vlasceanu, the lead researcher, explained that the assumption that AI ethics is a problem of engineers' knowledge or motivation is incorrect. Instead, the real issue lies in the working conditions within organizations. Engineers described environments where raising ethical concerns could harm their careers and where the work required for true ethical AI development is not recognized or rewarded. Vlasceanu emphasized that to build AI systems that are truly ethical, the focus should be on changing the conditions under which AI is developed, not just on training individual engineers more. At a time when governments and organizations are introducing new AI regulations, such as those under the EU AI Act, the study suggests that many current efforts focus on creating documents, policies, and reports to demonstrate ethical commitment. However, the research indicates that unless the actual development processes are changed, these measures may be little more than a formality. Professor Caroline Jay, who supervised the research, noted that the issue is not about bad companies or bad engineers, but about the gap between the appearance of ethical practice and the reality of it. She pointed out that the infrastructure designed to govern AI ethics has grown faster than the technology itself, and that this infrastructure is often aimed at the wrong level. The study argues that real change must focus on how AI projects operate on a daily basis—how ethical concerns are raised, who is responsible for addressing them, and whether thorough testing and safety work are recognized and rewarded. It also calls on regulators to go beyond checking whether the right documents have been created and instead ensure that ethical safeguards are actually being followed. This research is part of a broader doctoral project examining the relationship between engineers' intentions, public perceptions of AI, and the outcomes of AI systems. The researchers plan to test these findings with a larger survey of engineers and hope the work will help shape future AI governance by focusing on the organizational conditions that influence how AI systems are developed and used.