Pandora's Black or White Box: Are AI Tools Undermining Evidence?

The proliferation of synthetic media, particularly deepfakes, is having a significant impact on society in general and the criminal justice system in particular. The increase in deepfake audio and video evidence presents significant challenges to digital forensics, where accurate detection is essential for maintaining the integrity of evidence. With limited manual methods available for identifying manipulated content, forensic professionals are increasingly turning to AI-enabled detection tools. This research examines whether the application of professional-grade AI tools for deepfake detection introduces changes to original files, such as modifications to metadata, compression artefacts, or structural alterations, that could compromise their integrity and admissibility in legal contexts. The investigation has focused on 2 professional tools currently widely employed within audio and video forensic units of UK police forces. The work undertaken employs controlled testing to assess the extent and nature of such changes. It evaluates whether these tools adhere to ethical and procedural standards, particularly regarding the chain of custody and evidentiary reliability. The findings aim to inform the integration of AI tools into audio/video forensic workflows, addressing critical concerns about balancing advanced detection capabilities with the preservation of file authenticity. Through identifying potential risks and challenges, the study seeks to support policymakers, forensic practitioners, and legal professionals in ensuring that advancements in synthetic media detection uphold the foundational principles of evidence handling and judicial integrity.