Research on the impact of large language model turnover on screening for AI-assisted scientific writing, highlighting the need for re-verification of benchmark accuracy with every LLM release.