Overview
The rapid advancement of deepfake technologies is making manipulated media increasingly difficult to distinguish from authentic content. For organisations, this creates significant exposure to fraud, impersonation, operational disruption, reputational damage and loss of stakeholder trust.
DeepShield is an explainable, multimodal deepfake detection technology designed to help organisations assess the authenticity of digital media. It analyses multimedia content for signs of manipulation and provides localised evidence to support technical review, investigation and audit processes. Designed with Asian media and language contexts in mind, DeepShield can serve as a digital trust layer within fraud prevention, cybersecurity, content verification and compliance workflows.
Our Innovation
Many deepfake detection tools provide only a classification or probability score, offering limited insight into what influenced the result. DeepShield combines detection with explainable evidence, helping reviewers understand whether media may have been manipulated, where relevant indicators appear and how those indicators evolve over time.
The technology performs temporal and scene-level analysis across video sequences. Its research evaluation interface provides clip-level assessment signals, highlights selected high- and low-frequency evidence regions, and presents temporal-slice explanations of local pixel evolution. These interpretable outputs support informed human review and the creation of clearer investigation and audit records.
Technology Features and Advantages
- Explainable evidence: identifies localised regions and temporal signals that contribute to an assessment, supporting technical review and auditability.
- Temporal and scene-level analysis: Examines changes across video sequences rather than relying solely on isolated frames.
- Multimodal approach: Provides a foundation for analysing manipulated multimedia content across different information signals.
- Asian-context design: Developed with Asian media and language environments in mind, with potential for further regional and sector-specific adaptation.
- Compute-efficient architecture: Designed to support rapid analysis, with edge deployment, continuous learning and fast adaptation identified in the development roadmap.
- Integration potential: Envisioned for integration through APIs or edge-based deployment within existing enterprise and security workflows.
Potential Commercial Applications
- Identity, fraud and compliance: Supporting KYC, account recovery, claims verification, fraud investigations and regulated evidence review.
- Cybersecurity and digital platforms: Detecting suspected executive impersonation, manipulated content and deepfake-enabled social-engineering threats.
- Media and corporate communications: Assisting with source verification, publishing decisions and the validation of executive or organisational communications.
- Government and public services: Supporting scam response, investigations, citizen services and the verification of public-facing digital media.
- Digital forensics: Providing explainable analytical outputs to assist investigators in reviewing and documenting suspicious media.
Collaboration and Licensing Opportunities
SMU welcomes collaboration with organisations interested in evaluating DeepShield using relevant datasets and operational scenarios, conducting proof-of-concept projects, adapting the technology for specific industry requirements, or integrating its detection and explanation capabilities into existing products and workflows.
Opportunities are available for technology evaluation, research collaboration, product integration and licensing.
If you are interested in this technology, please contact SMU’s Knowledge Transfer and Commercialisation team.