Nodding, blinking, shaking your head - this set of actions is a fixed operation that many apps require users to perform during face verification. A typical case recently announced by the Hangzhou Intermediate People's Court has aroused users' concerns about facial recognition: two criminals used photos collected online, used AI to generate dynamic videos of nodding and blinking, and then illegally logged into personal accounts through virtual camera applications and other means.
In the first instance, the court held that the above-mentioned behavior infringed upon the personal information rights of many unspecified social subjects, harmed social and public interests in relevant fields, and constituted infringement. So how do criminals use AI-generated content to fool the identification system? In response to this situation, how should we take preventive measures to protect the security of personal information?
In the case announced by the Hangzhou Intermediate People's Court, in order to make profits, Zhang and Wang obtained other people's mobile phone numbers, avatar photos, identity information, etc. from an overseas social software through some means, and used "AI face-changing" technology to create fake liveness verification videos, forced them to log in to other people's shopping platform accounts, sold their delivery addresses and shopping cart information to third parties, and also paid for false real-name authentication for others.
Ye Yongqing, assistant judge of the Intellectual Property Tribunal of the Hangzhou Intermediate People's Court: Traditional infringement leaks information. In this case, the important line of defense of identity verification was broken. What the victims may face is no longer the leakage of a single piece of information, but a series of chain risks such as account takeover, property transfer, identity fraud, etc. The court's attitude towards this is also very clear. Biometric information should be protected under the strictest protection, and the evaluation of this type of behavior cannot stop at the level of individual losses.
Unlike identity information and photo leaks, identity verification breakthroughs are more difficult. A staff member of a large bank told reporters that the bank's remote face verification is not simply watching a dynamic video. In addition to actions such as opening the mouth and shaking the head in the screen, there are also multiple layers of anti-counterfeiting checks in the background. The system may detect the imaging noise, light and shadow consistency, texture, 3D depth information, etc. of the picture. In addition, it must pass the inspection of the living body detection algorithm. So how does AI-generated content fool the recognition system?
Zhang Xiaohan, a young associate researcher at the Development Research Institute of Fudan University, said that the forgery methods are mainly divided into two parts: facial information theft and live simulation.
In this case, Zhang and Wang used "AI face-changing" technology to identify the target person's photos and extract face information through deep learning algorithms, and then continuously used this information to train and model to achieve the effect of face-changing. In addition to making fake face authentication videos, the two also replaced the local camera by implanting a virtual camera application in the mobile phone. In this way, when the platform called the camera for face recognition verification, the face on the screen was automatically replaced with this synthetic video.
In the past, intrusion methods such as face-changing required a professional team, but now only open source tools and a mobile phone can be used to commit crimes, and AI infringement has shown a trend of lower thresholds. Are methods such as face authentication still reliable? Zhang Xiaohan said that there is a need to enhance personal information security awareness, but there is no need to panic too much.
Zhang Xiaohan: Our country has actually introduced a series of policies, regulations and national standards. But on the other hand, the ability to generate videos is indeed getting stronger and stronger, so for the public, on the one hand, personal privacy must be protected, and on the other hand, multi-factor authentication or two-factor authentication can be enabled in the App. In this way, in addition to verifying the face, it will also verify the mobile phone verification code, etc.
Ye Yongqing said that the legal aspects are still being updated to provide protection for the security of personal information.
Ye Yongqing: The Civil Code, Personal Information Protection Law and the newly revised Cybersecurity Law all have relevant provisions on facial information or biometric information. In addition, the Supreme People's Court also issued the "Opinions on the Legal Trial of Cases Involving Artificial Intelligence Disputes", further clarifying the judicial rules for cases involving artificial intelligence. Currently, our country is promoting the process of special legislation in the field of artificial intelligence.
