Automatic writing software has become a common tool in the creation of research articles, according to a study published in PNAS by Yongyuan He and Yi Bu from Peking University. Conducted in March 2026, the study analyzed over five thousand article titles and millions of research papers, finding that policies implemented by thousands of journals have not reduced the use of AI in writing. The rate of AI use in scientific publications remains consistent, regardless of whether a journal requires transparency or not. Since the introduction of ChatGPT at the end of 2022, the use of computer-assisted writing has grown rapidly across all academic fields. By examining 5,114 journals and 5.2 million articles from a large bibliographic database, the researchers assessed how editors have responded to the rise of AI. Nearly 70% of journals have adopted AI-related policies, most commonly requiring authors to declare their use of AI, though some journals have imposed outright bans on certain tasks involving AI. Despite the increase in such policies, the number of articles that actually declare AI use remains extremely low. Out of about 75,000 recently analyzed articles, only 76 explicitly mentioned the use of AI, or about 0.1%. While this percentage has increased from around 0.1% in 2023 to 0.43% in early 2025, it is still far below the actual usage detected in the text. To identify AI-generated content, the researchers looked for specific language patterns, such as the overuse of certain generic words. One of the most surprising findings was that AI use has increased at the same rate in journals with strict policies as in those without any rules. This suggests that transparency policies or even outright bans on AI use have little impact on actual behavior. The trend is especially noticeable in open-access journals and in the physical sciences, where researchers from non-English-speaking countries often use AI tools to overcome language barriers. For these researchers, AI software acts as both a proofreader and a translator, helping them meet the growing demand for English-language publications. The study highlights a growing concern: voluntary declarations of AI use are no longer enough to ensure the traceability of scientific research. Alex Glynn, from the University of Louisville and a co-author of the study, argues that failing to disclose AI use is essentially equivalent to lying. As a result, discussions are shifting toward the need for stricter rules, whether to improve the detection of AI-generated texts or to revise ethical guidelines. He and Bu specifically call for a more precise understanding of AI's role in scientific research. Without such transparency, suspicions of fraud may continue to erode trust in research that is becoming increasingly difficult to verify.