ישראל מירסקי

Fingerprint biometrics are extensively used for identification and security, from border control to consumer electronics. The permanence of fingerprints means security breaches can have lasting impacts. Traditionally, organizations store only fingerprint templates to mitigate the risks associated with stolen fingerprint images. However, advances in generative artificial intelligence (GenAI), particularly deepfake technologies, now enable adversaries to generate fingerprints from stolen templates, increasing the threat of data breaches. In this paper, we demonstrate and validate a previously theorized threat by evaluating an end-to-end attack in the physical world. Our approach involves: (1) generating fingerprint images from unseen templates, (2) fabricating silicone replicas of these deepfake fingerprints using a 3D resin printer, and (3) successfully deceiving fingerprint scanners with the replicas. The entire lab setup cost only $440 USD and only 7 cents to replicate each fingerprint thereafter, highlighting the attack’s practicality. To support reproducibility and encourage the development of defenses, we publicly release our deepfake fingerprint pipeline.