This study examined auditors' trust in AI-based audit tools that function as "black box" systems in various Public Accounting Firms (KAPs) in East Java, Indonesia. Based on theories of trust in technology and technology adoption, this study proposes that auditor trust is shaped by cognitive trust in reliability and explainability, social norms within the audit team, and the organizational structure that guides AI implementation. A qualitative approach using the Interpretative Phenomenological Analysis (IPA) methodology was chosen, using in-depth interviews with auditors from small, medium, and internationally affiliated firms to investigate how they interpret, negotiate, and even reject AI output recommendations in their audit practices. The findings indicate that auditors often place conditional trust in AI output. They position AI tools as mere tools that support traditional audit procedures and professional judgment. The higher the auditor's level, the lower the level of trust in AI. AI adoption by public accounting firms is influenced by their business models, strategic risks, and reputations. This study also reveals infrastructure gaps, differences in AI access capabilities, and a strong relational culture in AI integration. This study advances the literature on AI in auditing by presenting an empirically grounded conceptual model of the development of auditor trust in AI-based audit tools operating in a "black box.".