Generative artificial intelligence has rapidly entered higher education writing classrooms, yet many institutions remain uncertain about whether AI feedback should be prohibited, tolerated, or deliberately integrated into pedagogy. This study examines a responsible AI feedback model in first-year academic writing courses by comparing three feedback conditions: instructor-only feedback, AI-assisted feedback, and a guided hybrid model combining AI feedback with instructor mediation and explicit integrity instruction. Using a quasi-experimental mixed-methods design, the study followed 126 first-year students across three parallel writing course sections over an eight-week essay revision unit. Quantitative data included pre- and post-revision writing scores, a self-regulated learning scale, a feedback uptake index, and an academic integrity clarity scale. Qualitative data were drawn from student reflection logs, prompt records, and semi-structured interviews with 24 students. The simulated results indicated that the guided hybrid group showed the strongest improvement in revision quality, with an adjusted posttest mean of 82.6 compared with 76.8 in the instructor-only group and 78.4 in the AI-assisted group. The hybrid group also demonstrated higher feedback uptake, stronger monitoring of revision decisions, and clearer understanding of acceptable AI use. The findings suggest that AI feedback is most educationally valuable when embedded in transparent, human-supervised, feedback-literate pedagogy rather than used as an autonomous writing correction tool.