BioID GmbH

98% of deepfakes detected with the official HydraFake dataset Regular benchmarking is part of our ongoing work on liveness detection (PAD) and deepfakes. HydraFake-100K is a deepfake benchmark comprising 100,000 images - 50,000 real and 50,000 fake - sourced from a variety of places. It is designed to evaluate the ability of deepfake detectors to generalise to unseen generation models, forgery techniques and image domains. The result of our internal testing is encouraging: approximately 98% of the tested fake samples were correctly classified as fake, while the test data, needless to say, has not been used for training. This is an encouraging finding in our ongoing work to strengthen biometric security. Interested in more details? Let’s connect! #Biometrics #Deepfakes #DigitalIdentity #Cybersecurity #AI #HydraFake

A face can match. And the identity verification can still be wrong. As AI-generated faces and real-time face swaps become more sophisticated, liveness detection has moved from an optional add-on to an essential defence. The problem's scope is rapidly expanding. In 2023, there were about 500,000 voice and video deepfakes on social media; by 2025, that figure is predicted to rise to eight million. Attackers can now generate forgeries in minutes as opposed to weeks in 2022. (Fintech Global) Organisations need to distinguish between the following: - Face matching: does the face correspond to the claimed identity? - Liveness detection: is there evidence that a real person is present rather than a photo, replay, or other spoof? - Injection-attack detection: has manipulated biometric data been inserted into the digital capture or verification pipeline? These controls address different risks. Treating them as interchangeable can leave important gaps in remote onboarding. For banks, fintechs and digital service providers, the challenge is to combine strong security with a smooth customer experience. How is your organisation adapting its identity verification process to AI-enabled fraud? #LivenessDetection #Deepfake #KYC #FaceBiometrics #FraudPrevention

"D-you" is coming. Your identity is going digital. But how do we know it's really you? Germany's EUDI Wallet now has a name: d-you. The German digital wallet is planned to launch in January 2027 with 40 partners from business, academia and public administration, bringing the European Digital Identity Wallet one step closer to everyday use. For citizens, this promises a simpler way to use verified digital credentials while maintaining control over their personal data. For organisations, it raises another important question: How do we know that the person using a digital identity is actually the person to whom it belongs? This is where biometrics play an important role. Identity credentials can establish what information is associated with an identity. Biometric verification can add assurance that a real person is present and that they are the legitimate holder of that identity. At BioID, we use a multi-layer approach combining liveness (PAD), deepfake detection, and challenge response to identify manipulated or AI-generated content, strengthening trust throughout the digital identity lifecycle and combating injection attacks. Presence → Is a real human there? Authenticity → Is the video genuine? The EUDI Wallet is about giving people more control over their digital identity. The next challenge is ensuring digital identity and real-world identity remain securely connected. What role do you think biometrics should play in the EUDI Wallet ecosystem? #EUDIWallet #dYou #eIDAS #DigitalIdentity #Biometrics #DigitalTrust #IdentityVerification #LivenessDetection #EuropeanDigitalIdentity

Deepfake detection is an evolving field. As generative models improve, detectors must continuously adapt. This highlights the need for security and usability to evolve together. Read our latest newsletter to understand why we shouldn't base security solely on today's AI models. Sora may be obsolete, and Seedance may be the current benchmark. Tomorrow, another model will emerge to take its place, and that is precisely the point.

Biometrics Made in Germany – discussed this week at the annual Darmstadt Biometrics Week, hosted by Fraunhofer IGD. With an agenda full of #injection attack and #deepfake detection, this is just the right place to be for our team. Attending, presenting, demonstrating – as every year for decades, we are excited to support and learn. Biometrics is no longer only about what happens in front of the camera – this is our takeaway from the German @IT Security Association Germany (TeleTrusT) Biometrics Working Group workshop, organised in cooperation with the European Association of Biometrics (EAB) at Fraunhofer IGD. One of the highlights was the discussion around injection attacks, deepfakes and, particularly, how the attack surface is evolving. For years, much of the conversation around biometric attacks focused on what could be presented to the sensor: a photograph, mask or other physical artefacts. Injection attacks change the picture. An attack can take place after the biometric sensor and before signal processing, where biometric data or signals can potentially be manipulated or injected digitally. And this has an important consequence: Scalability. Unlike many physical presentation attacks, digital injection attacks can potentially be automated and repeated at scale across applications, devices and environments. The deepfake landscape makes this even more relevant. From face swaps and facial attribute editing to synthetic identities, face morphing and generated video, the ability to create increasingly convincing biometric content continues to evolve. We appreciated observing the evolution of terminology, standards, and threat models surrounding attack vectors that, from a technological standpoint, intersect with challenges the biometrics industry has been tackling for years under various names. The takeaway is simple: Biometric security cannot stop at asking whether the face looks real. We increasingly need to ask whether the entire biometric signal and processing pipeline can be trusted. This is where liveness detection, injection attack detection and deepfake detection increasingly come together. A very worthwhile day in Darmstadt, and another reminder that in digital identity, the attack surface keeps moving. #Biometrics #InjectionAttacks #Deepfakes #DigitalIdentity #LivenessDetection #CyberSecurity #Identity #TeleTrusT #EAB #Fraunhofer"

The ID Is Legitimate. But Is the Person Real? A deepfake is the content. An injection attack is the delivery method. Did you know that an attacker can potentially inject synthetic media directly into the processing pipeline without presenting it to a physical camera at all? Presentation Attack ↓ Fake content is presented to the camera. vs. Injection Attack ↓ Fake content is inserted in the biometric process, skipping/manipulating the real camera signal. BioID helps protect across these attack surfaces. 1️⃣ BioID's Face Liveness Detection is independently tested to ISO/IEC 30107-3 Level 3 / Level C for sophisticated presentation attacks. 2️⃣ BioID's injection attack detection (#IAD) helps, if a process is attacked differently - by injecting media directly, suggesting the photo or video is being captured live by the camera but instead using prerecorded or manipulated material. 3️⃣ Complementary AI-powered #deepfake detection helps identify synthetic or manipulated facial content - no matter if presentedto the camera or injected in the process. Interested? Test our tech at the BioID Playground! #bioid #identityfraud #deepfakedetection #injectionattack #deepfakes #biometrics #liveness