Incode partners with Purdue University for deepfake detection research

A convincing deepfake now takes minutes to create and costs almost nothing to distribute. Generative AI is producing synthetic media at an unparalleled speed, and the rise of agentic fraud is expanding these capabilities faster than detection methods can keep pace. To truly overtake fraudulent actors, sustained research and collaboration are critical.

Today, we are announcing a research collaboration between Incode and Purdue University that aims to address the gap between deepfake creation and detection.

In this post, you will learn why credible deepfake defense depends on collaboration between industry and academia, what the two teams will focus on, and where the research partnership goes from here.

Why collaboration between industry and academia is essential

Production identity systems and academic labs approach the deepfake problem from different vantage points, and each observes phenomena that the other misses.

Companies operating identity verification at scale encounter real attacks every day. Face swaps, injection attacks, and synthetic documents show up in live traffic long before they appear in public research datasets. That exposure keeps detection work grounded in how fraud behaves in the field, and it surfaces new techniques while they are still emerging.

Meanwhile, academic researchers contribute a different strength. University labs bring scientific rigor, peer review, and institutional independence, along with the freedom to study challenging problems over years instead of quarters. Their work sets the bar for how detection methods should be measured, compared, and trusted.

Credible deepfake defense requires both approaches. Detection trained on lab data alone misses what attackers are learning and iterating on in the moment, while detection tuned solely to production traffic can lack independent scrutiny and rigor. Durable progress comes from pairing the two. That is exactly what the partnership between Incode and Purdue University is designed to do.

Inside the Incode and Purdue University partnership

Incode and Purdue University are collaborating to advance deepfake detection and improve how detection systems are evaluated in real-world identity verification. The partnership pairs Incode’s machine learning team with Purdue researchers on five shared goals:

• Improve detection of new and previously unseen synthetic media
• Develop more rigorous and credible approaches to independent evaluation
• Connect academic research with real production conditions
• Advance responsible practices around biometric data, privacy, and research governance
• Publish research and resources that help move the broader field forward

Incode’s track record in deepfake detection

Incode brings to the collaboration an active research program in synthetic media and presentation attack detection (PAD).

In 2026, Incode won both tracks of the IJCB PAD-ID Card document forgery detection competition and took first place in the video category of LivDet-Face 2026, independent benchmarks presented at the International Joint Conference on Biometrics. Incode is the only team to win both PAD-ID Card tracks in the competition’s three-year history.

Competition results like these matter because independent organizers run the judging. Every entrant faces the same sequestered data (test sets the participants never see) under the same conditions, so the results reflect the models rather than the marketing.

The same research feeds Deepsight, Incode’s deepfake and injection-attack detection system, which analyzes biometric, behavioral, and device signals in real time to confirm a real, live person is present.

"We see new synthetic media techniques in production before most of the field knows they exist. Working with Purdue lets us pair that visibility with the scientific rigor this problem demands.” — Efim Boieru, Senior Manager of Machine Learning at Incode

Purdue’s track record in media forensics

The collaboration draws on the Purdue Machine Learning and Media Forensics (M2) Lab, directed by Prof. Shu Hu of Purdue’s School of Applied and Creative Computing. The lab specializes in media forensics, the science of determining whether digital images, video, and audio are genuine or manipulated, along with the fairness and reliability of the AI models that make those calls.

The lab’s recent work spans some of the most challenging problems in the field: keeping deepfake detection accurate and fair across demographic groups, building a million-scale dataset of AI-generated faces to benchmark detectors, and detecting AI-synthesized voices. That research has appeared at leading peer-reviewed venues, including CVPR, one of the world’s top computer vision conferences.

“Deepfake detection research requires rigorous testing. This partnership connects our lab to the conditions detection systems face in the real world, helping us build evaluation methods that hold up to bad actors in the field.” — Prof. Shu Hu, director of the Purdue Machine Learning and Media Forensics (M2) Lab

What’s next for the partnership

The Incode and Purdue University collaboration is built for the long term. Joint work will focus on the challenges that will define the next several years of deepfake defense: detecting which generation techniques will come next, building evaluation frameworks the field can rely on, and understanding how AI-driven identity fraud evolves as agents and automation spread.

Both teams plan to publish findings and resources openly when they are ready, so the research strengthens the broader community working on synthetic media defense.

A more credible standard for deepfake defense

Generative AI will keep improving, and the attacks built upon it will improve alongside it. Addressing this issue requires a defense that advances on the same curve, grounded in production experience and proven through independent, scientific evaluation. That is the standard this partnership sets out to build.