One of the key advantages of open-access journals is that they make scientific research freely available to anyone with an internet connection, removing financial barriers and expanding the reach of knowledge. However, the rapid growth of this publishing model has also enabled the proliferation of questionable or predatory journals. These outlets often charge authors publication fees but offer little to no peer review, fast-track publishing, and minimal editorial oversight. To combat this issue, researchers have developed a new AI tool capable of identifying red flags in open-access journals, helping scientists avoid submitting their work to disreputable publishers. In a study published in Science Advances, the team describes how they trained an AI system to detect signs of low-quality or deceptive journals by analyzing over 12,000 reputable publications and nearly 2,500 journals that were previously removed from the Directory of Open Access Journals (DOAJ) for violating standards. The AI was taught to recognize warning signs such as missing or vague information about editorial boards, unprofessional website design, inconsistent formatting, and unusually low citation rates. After training, the model was applied to a massive dataset of 93,804 open-access journals sourced from Unpaywall, a platform that helps users locate free versions of paywalled research. The results were striking: the AI flagged more than 1,000 previously unknown journals that collectively publish hundreds of thousands of articles. Many of these suspect journals appear to originate from developing countries, though the study does not name specific outlets, citing concerns over potential legal consequences. While the tool demonstrates strong potential for large-scale screening of journals, it is not perfect. The system currently has a false positive rate of 24%, meaning it incorrectly identifies about one in four legitimate journals as problematic. As the researchers note, this underscores the importance of human expertise in verifying findings. They emphasize that AI should be used as a first-line screening tool, not a replacement for expert judgment. “Our findings demonstrate AI's potential for scalable integrity checks, while also highlighting the need to pair automated triage with expert review,” the authors write. The battle to maintain scientific integrity in scholarly publishing is ongoing. As predatory publishers evolve their tactics, so too must the tools used to detect them. The researchers believe future improvements to the AI—such as refining its feature set and adapting to new patterns—could make it even more accurate and reliable. Ultimately, the most effective defense lies in combining the speed and scale of artificial intelligence with the discernment and experience of human experts. Together, they can help protect the credibility of science and guide researchers toward trustworthy publishing venues around the world.