Binghamton University uses physical sensor markers for deepfake detection methods

This shift toward physical forensics addresses the diminishing returns of software-only detection as generative AI reaches visual parity with real footage. By focusing on hardware-level artifacts like sensor fingerprints and electrical mains hum, researchers are creating a verification layer that AI models cannot easily replicate because they bypass the physical recording process entirely. For the streaming and media ecosystem, this provides a more durable defense against sophisticated misinformation and synthetic content. As these methods mature, the industry should watch for the integration of these physical verification tools into standard content authentication workflows and digital forensic toolkits used by major platforms.

Binghamton University's CAITI cluster builds on years of prior work by Professor Yu Chen's lab in physical-anchor verification. In early 2025, Chen won $50,000 from the SUNY Technology Accelerator Fund to develop CerVaLens, a lightweight deepfake detector designed for mobile devices including Google Pixel 10 smartphones, with the stated goal of moving the technology from controlled lab environments into real-world deployment for law enforcement, digital forensics, and intelligence operations. The TAF grant specifically funds testing against hundreds of thousands of images, videos, and audio samples generated by popular AI tools, a scale-up that directly precedes the CAITI cluster's broader mission. The technical foundation CAITI now consolidates includes multiple named tools from Chen's group. The team published research in the Proceedings of Disruptive Technologies in Information Sciences comparing real and AI-generated images using frequency-domain signal processing, alongside DeFakePro, a tool that verifies recordings by analyzing electrical network frequency signals embedded in media files. DeFakePro's ENF-based approach is the direct predecessor of the electrical grid marker methodology that CAITI now formalizes as a research program. Virginia State University, which signed a partnership statement at the CAITI launch, has already collaborated with Chen's lab on these publications, with Professor Enoch Solomon and master's student Monica Sudarsan co-authoring the GANIA and DeFakePro work. The CAITI cluster's emphasis on sensor forensics also connects to longer-standing research at Binghamton. SUNY Distinguished Professor Jessica Fridrich, who presented at the cluster launch, is recognized as a pioneer of camera-sensor fingerprint detection for digital forensics, a technique that identifies the specific device that captured an image based on manufacturing imperfections unique to each sensor. Binghamton's earlier collaboration with Intel produced FakeCatcher, a deepfake detection tool that achieved over 90% accuracy by analyzing photoplethysmography signals, or subtle blood-flow patterns visible in video. FakeCatcher demonstrated that physiological signals embedded in real video provide a detection axis that generative models do not replicate, a principle CAITI now extends to electrical and hardware-level physical anchors. Recent industry developments, such as Google SynthID Detector opens to public for AI video verification, underscore the growing demand for these robust verification methods.