Location, blood flow research show deepfake problem’s increasing sophistication

The growing volume of deepfake fraud and increasing sophistication of the technology is forcing researchers to consider how to confirm the authenticity of content and the metadata that goes along with it. That means finding a way to verify claims about where images come from, and when, but also considering the finer details of characteristics like the subtle changes in facial coloring from blood flow. If AI mimics these changes, does it do so effectively, or can its mimicry still be detected?

Decentralized system to cryptographically verify place and time

In a world awash with synthetic media and GPS spoofing, proving that something truly happened in a specific place at a specific time has become increasingly fraught. Eduardo Brito, an intersectoral doctoral researcher at the University of Tartu Institute of Computer Science, believes he has an answer.

His newly published study “Decentralized Proof-of-Location systems for trust, scalability, and privacy in digital societies” in Scientific Reports lays out the first unified, system-level architecture for a decentralized Proof-of-Location (PoL) system for images that can cryptographically validate physical presence without exposing private data or depending on central authorities.

“Imagine someone says, ‘I was at this place at this time.’ PoL makes it possible to verify that claim without simply trusting their word or revealing unnecessary private data,” Brito explains. Rather than relying on vulnerable GPS signals or centralized databases, his model recruits nearby devices, called witnesses, to co-sign location claims. Each witness cryptographically attests to the presence of a device (the prover) at a given time and place, creating a digital stamp that third parties can later verify.

Brito and his colleagues at Cybernetica AS tested their concept by filming volunteers wearing smartwatches and heart monitors under varied lighting and motion conditions. They then compared recorded heart rates against color shifts detected at 79 facial points, demonstrating that advanced sensors and algorithms can indeed corroborate physical presence. Early results were described as “very promising,” paving the way for broader applications.

Potential uses span from authenticating the origin of photos and videos to securing supply chains, supporting digital alibis in legal disputes, and even enabling anonymous, verifiable attendance in civic processes like voting. Unlike centralized alternatives, the decentralized PoL framework builds in safeguards against mass surveillance and involuntary witnessing. Devices must cryptographically consent before acting as witnesses, and proofs expose only the minimal information required for verification.

Still, Brito acknowledges that adoption will not be straightforward, noting that legacy systems are deeply embedded and expensive to replace. Entities that profit from controlling location data, whether large tech firms or surveillance-heavy governments, may resist a model that strips them of gatekeeping power.

He expects early uptake in areas where location fraud already poses serious risks, such as content authentication and logistics, before expanding into broader civic and infrastructure uses as regulations catch up.

Looking ahead, the research team is scaling prototypes and exploring pilot deployments with governments, NGOs and private companies interested in journalism, digital asset verification and critical infrastructure monitoring. Open sourcing the protocol is part of their roadmap, ensuring that transparency and community audit remain at the project’s heart.

Brito envisions that within five years, Proof-of-Location could gain acceptance as legal evidence and in emergencies, from natural disasters to armed conflicts, to provide rapid, trustworthy confirmation of events on the ground.

In a landscape rife with digital deception, this decentralized approach to verifying presence could be a cryptographic foundation for truth, the researchers suggest.

Face color analysis to combat deepfakes

The Netherlands Forensic Institute (NFI) has developed a novel way to spot deepfake videos by tracking the tiny shifts in facial color caused by a person’s heartbeat.

Dubbed “blood flow detection,” the technique harnesses advanced image analysis to pick up on the subtle skin‐tone variations driven by the pulse. It was presented prior to the research’s publication at the European Academy of Forensic Science (EAFS 2025) conference in Dublin, Ireland at the end of May.

Zeno Geradts, a digital forensic investigator at the NFI and professor of forensic data science at Amsterdam University, warns that the rush of synthetic content makes this work more urgent than ever.

Though blood flow detection isn’t yet fully validated for courtroom use, investigators can already deploy it to support individual inquiries. With the NFI forecasting a surge in demand for deepfake analysis, the new tool arrives at a critical moment.

Artificial intelligence can do a lot, but it still cannot generate a convincing pulse,” Geradts said (via Dutch News). “In real video, you can detect blood flow around the eyes, forehead and jaw — that’s what’s missing in deepfakes.”

In recent trials, volunteers wore smartwatches and heart monitors while being filmed under varying light and motion conditions. Analysts compared the recorded heart rates against colour changes at 79 facial points. The early results, Geradts reports, are very promising.

However, a paper published in April in the journal Frontiers in Imaging from researchers with the Fraunhofer Institute for Telecommunications and Humboldt University, both in Berlin, point to the increasing realism of deepfakes. The researchers found that generative AI can now replicate the variations caused by the pulse.

The “High-quality deepfakes have a heart!” researchers used remote photoplethysmography (rPPG) to analyze local blood volume changes, and found that the method produces reasonable estimates of pulse rate, as measured by the ECG. But they also found that when inserting digitally altered faces the deepfakes retained the effect of heartbeat in the videos.

Analysis of the spatial distribution of blood-flow, however, could still be useful for detecting these high-quality deepfakes.

“The developments are moving so fast,” Geradts said in an NFI post on the research earlier this year. “But I haven’t seen any deepfakes where the heartbeat is perceptible to the face. Are they going to come after publication of our research? It could be. But we can certainly move forward for this over the next two years. There are no tools in this field for eternity.”

Article Topics

deepfake detection  |  deepfakes  |  digital trust  |  heartbeat biometrics  |  location data  |  synthetic data