This study examined the impact of source information indicating AI authorship on writing assessment and its effects on overall scores, text quality levels, and analytic criteria. Using a repeated-measures design, 20 Korean L1 teachers evaluated identical AI-generated texts, with and without AI authorship information. The results showed a significant decrease in scores when AI authorship was disclosed, particularly for mid-level texts and criteria involving interpretive judgment (e.g., ideas and supporting evidence; sentence and vocabulary). The qualitative findings suggest that the raters applied higher expectations and critically reinterpreted the content in terms of depth and authenticity. Thus, writing assessment is shaped by rater effects, framing effects, and algorithm aversion. This study has implications for fairness and validity in AI-era writing assessment.