This mixed-methods study investigated the ethical landscape of AI-assisted academic writing and examined how philosophical positions on machine agency influenced ethical attitudes and behavioral intentions among East African academics and students. Employing a sequential explanatory design, the research collected data from 847 participants (521 students, 326 faculty) across diverse disciplines through structured questionnaires measuring AI usage patterns, ethical attitudes, and philosophical beliefs about machine creativity, intentionality, and moral status, supplemented by 32 semi-structured interviews. Univariate analyses revealed moderate-to-high AI adoption (M = 3.59, SD = 1.01) accompanied by moderate ethical concerns (M = 3.34, SD = 0.84), with significant variation across participant type, academic discipline, career stage, and familiarity level. Bivariate analyses demonstrated that students reported significantly higher usage and lower concerns than faculty (t = 6.24, p < .001), engineering scholars showed the highest acceptance while humanities scholars exhibited the greatest skepticism (χ² = 47.32, p < .001), and AI familiarity inversely predicted ethical concerns (F = 89.45, p < .001). Correlation analyses revealed strong intercorrelations among machine agency indicators (r = .58 to .71) and substantial relationships between machine agency beliefs and ethical acceptability (r = .56, p < .001), disclosure necessity (r = -.39, p < .001), and integrity concerns (r = -.49, p < .001). Structural equation modeling confirmed excellent model fit (CFI = .956, RMSEA = .036) and demonstrated that machine agency beliefs significantly predicted ethical acceptability (β = .487, p < .001), disclosure attitudes (β = -.312, p < .001), and integrity concerns (β = -.394, p < .001), with academic discipline serving as a significant moderator such that the agency-acceptability relationship was attenuated in humanities (β = -.216, p = .001) and amplified in engineering (β = .198, p = .002). The model explained 38.9% of variance in ethical acceptability, 26.7% in integrity concerns, and 15.6% in disclosure necessity. Qualitative findings revealed that participants struggled to distinguish legitimate AI assistance from problematic substitution, often relying on intuitive rather than principled judgments, and expressed desire for clear institutional guidance that acknowledged disciplinary differences while maintaining scholarly integrity. The study concluded that philosophical positions on machine agency served as foundational premises from which practical ethical judgments were derived, yet these positions paradoxically reduced transparency impulses when AI was attributed greater creative capacities. Evidence-based recommendations included implementing discipline-specific ethical guidelines with mandatory disclosure frameworks, developing comprehensive educational programs addressing philosophy of authorship and AI ethics, and establishing institutional mechanisms for continuous policy evaluation and adaptation. These findings contributed to theoretical understanding of how philosophical commitments shape technological ethics while providing practical guidance for institutions navigating the integration of AI tools into academic practices in ways that preserve core scholarly values of originality, integrity, and intellectual growth.