This article proposes a privacy-aware framework for AI-assisted teaching to address the tension between effective AI use and controllable privacy risk in teaching English for international communication. Based on the 12 week course of “English International Communication”, a teacher feedback group, a group using cloud-based GenAI directly, and a privacy protection AI group were set up to organize teaching around four tasks: international email negotiation, policy/news oral briefings, cross-cultural forum hosting, and public service communication proposals. In terms of methodology, the system first performs local desensitization, policy gating, and task encoding on text, speech, and interaction logs, and then combines with the teacher anchor library to generate feedback. The privacy cost is measured by disclosure rate, re-identification risk, and feedback latency. The results showed that the post test comprehensive performance of the privacy preserving AI group was 84.9, higher than the teacher feedback group’s 74.6 and the direct cloud group’s 80.8; Its disclosure rate is 2.1%, significantly lower than the 8.7% of the direct cloud group, and the feedback latency remains at 7.2 minutes. Audience adaptation, cross-cultural appropriateness, and adoption rate of revisions all show a consistent improvement. Research shows that privacy protection entering the teaching process does not necessarily weaken the teaching value of AI; hen teacher calibration, task constraints, and local data processing are integrated into the same closed loop, artificial intelligence can form a more robust classroom support mechanism in international communication English teaching.
