Systematic review maps the state of AI deepfake detection across modalities and domains

Artificial intelligence based generative models have advanced rapidly, making it possible to produce highly realistic synthetic media. A new systematic literature review published in Artificial Intelligence Review examines how researchers are responding to the risks this creates, particularly the spread of deepfakes, which the authors describe as harmful synthetic content that can drive political misinformation, non-consensual imagery and financial fraud that costs millions. The review argues that these consequences threaten democracy, healthcare and the legal system, and it singles out an emerging concern: fabricated and manipulated diagnostic scans and clinical imagery that directly endanger patient safety, healthcare integrity and medical forensics.

The paper, led by Nitesh Budhwar and Pooja Mittal with colleagues including Sukanta Kumar Swain, Asisa Kumar Panigrahy, Vakkalakula Bharath Sreenivasulu and Yogesh Kumar Sharma, follows the PRISMA framework, the standard protocol for systematic reviews. The team searched major scientific databases including IEEE Xplore, Scopus, ScienceDirect, SpringerLink, Google Scholar and the ACM Digital Library. After applying predefined inclusion, exclusion and quality-assessment criteria, they selected 142 primary studies from an initial pool of candidates. Literature published before 2021 is cited separately for contextual background and is not counted among the 142 primary studies. The article carries a permanent DOI and is available as open access.

Once the corpus was assembled, the authors categorised the selected literature into different deepfake generation and detection methods. The review also analyses benchmark datasets, evaluation protocols and comparative performance trends across the field, giving readers a structured picture of which approaches dominate the literature and how they are measured.

The analysis identifies recurring weaknesses that limit real-world reliability. Dataset bias, poor generalization, adversarial vulnerability and a lack of explainability are discussed as key challenges facing deepfake detection systems.

Fabrication and manipulation of diagnostic scans and clinical imagery, the authors write, directly threaten patient safety, healthcare integrity and medical forensics.

Beyond cataloguing the present state of the field, the review outlines future research directions intended to guide work on more dependable detectors. It aims to serve as a useful reference for researchers and practitioners building reliable deepfake detection systems, consolidating a fast-moving literature that is otherwise scattered across multiple venues and databases. The authors frame the work as a response to a dual reality: the same technologies that create new opportunities for creative and legitimate use in various domains also give rise to significant concerns digital trust and privacy.

The 142 selected studies provide the evidentiary base for the categorisation of generation and detection methods, the dataset analysis and the challenge discussion.

Subject of Research: A systematic literature review of AI-based deepfake generation and detection methods across modalities and domains

Article Title: Building the trust: AI-based systematic literature review of deepfake detection across modalities and domains

Article References: Budhwar, N., Mittal, P., Swain, S. K., Panigrahy, A. K., Sreenivasulu, V. B., & Sharma, Y. K. (2026). Building the trust: AI-based systematic literature review of deepfake detection across modalities and domains. Artificial Intelligence Review .

Image Credits: AI Generated

DOI: 10.1007/s10462-026-11742-z

Keywords: deepfake detection, systematic literature review, PRISMA, generative models, synthetic media, medical imaging, dataset bias, adversarial vulnerability, explainability, benchmark datasets, digital trust, AI forensics

News Source: Denise Maddox. (November 6, 2026). Systematic review maps the state of AI deepfake detection across modalities and domains. Scienmag.