A facial recognition session at RSA Conference 2026 drew significant attention for moving past theoretical attack scenarios and demonstrating live, working exploits against production systems. The session, “Facing Reality: Hacking Facial Recognition,” was delivered Thursday afternoon by Jake Moore, a global cybersecurity advisor at ESET with a background in UK police digital forensics, and covered three distinct experiments that exposed practical weaknesses in deployed systems.
In the first demonstration, Moore used a pair of modified off-the-shelf smart glasses to identify people in real time while walking through a public space. Faces were captured and matched against publicly available data sources, with names and social media profiles returned within seconds. The experiment required no specialized hardware beyond commercially available eyewear and free software.
The second demonstration targeted financial services. Using AI-generated images and freely available software, Moore created a synthetic face and used it to open a real bank account through the institution’s facial recognition and eKYC onboarding flow. The system accepted the fictitious identity as genuine. Moore subsequently closed the account and disclosed the method to the bank, which has since closed that specific vulnerability. He noted, however, that the underlying question of how many other institutions remain susceptible is unresolved.
The third experiment went after law enforcement systems. Moore added himself to a facial recognition watchlist at a monitored train station in London, then walked through the area while running real-time face-swap software that overlaid a celebrity’s likeness onto his face in the camera feed. The system, which Moore said is used by UK police, did not flag him. “It is assumed that the camera feed is real,” Moore said. “Systems trust what they see on the screen, and so does the software.”
The experiments combined two pressures the biometrics industry has mostly treated separately: the ease of ambient face harvesting through wearable cameras and the effectiveness of AI-generated or AI-manipulated imagery against deployed liveness and matching systems. US senators recently asked Meta for answers about facial recognition plans for smart glasses, and EPIC has asked the FTC to investigate those plans. On the vendor side, the industry has also been shifting attention from presentation attacks toward injection-based threats.
Moore’s conclusion was direct: “The best way to verify someone is to bring them to another platform and communicate with them.” The implication for operators is that camera-based verification, without additional trust signals, may not meet the assurance bar that deployments in finance, law enforcement and border control are implicitly claiming.
Sources: WeLiveSecurity, ESET, RSA Conference
–
By Ali Nassar-Smith
