The present study examines how accurately Korean EFL teachers identify the authorship of English essays as student-written or AI-generated and explores the criteria they employ in making these judgments. It further investigates whether teachers’ perceptions of authorship influence their assessment of writing quality. The participants were 78 Korean EFL pre-service and in-service teachers, who evaluated a total of 12 essays for both authorship identification and writing quality. The results showed that teachers’ overall authorship identification accuracy was approximately 58%, only slightly above chance level, despite their relatively high level of confidence (7.13 out of 10 points). Teachers tended to associate student-written texts with limited idea development, personal and experience-based examples, and a limited vocabulary and syntactic variation, while attributing well-developed and information-rich ideas, consistent but formulaic organization, generalized content, and sophisticated vocabulary use to AI-generated texts. In addition, teachers’ perceptions of AI involvement were found to significantly influence their writing assessments. Given the limited reliability of authorship identification and the influence of perceived authorship on writing assessment, the study suggests a need to revise assessment rubrics to foreground writing process, idea development, and learner agency, while incorporating AI as a pedagogical scaffold in L2 writing instruction.