This study explored EFL university lecturers’ perceptions of generative AI in relation to critical thinking, focusing on perceived opportunities, acceptable-use boundaries, and assessment dilemmas at the University of Zawia, Libya. A mixed-methods descriptive design was used, combining a questionnaire (N = 70) and semi-structured interviews (n = 10). Survey results showed moderate endorsement of AI opportunities for supporting critical thinking (M = 3.93, SD = 1.08), especially for language and comprehension support (M = 4.23, SD = 0.97) and higher-order question generation (M = 4.03, SD = 1.05). However, lecturers strongly emphasized boundaries for acceptable AI use (M = 4.42, SD = 0.89), particularly the need for an institutional policy (M = 4.60, SD = 0.73) and disclosure of AI use (M = 4.49, SD = 0.86). Assessment dilemmas were high (M = 4.17, SD = 0.98), with strongest concern about plagiarism/patchwriting risks (M = 4.40, SD = 0.84) and difficulty judging students’ own critical thinking (M = 4.24, SD = 0.97). Interview findings reinforced a “benefit–boundary tension,” highlighting preference for assessment redesign (in-class writing, oral defense, and process portfolios) over detection-based policing. The study recommends regulated-use guidance, staff training, and student integrity support.