Network security and AI, cloud computing faults?

The annual GeekPwn International Security Geek Contest is held in Shanghai. In the past two years, the attack on AI is the most eye-catching of all projects, and the risks of cloud computing are equally compelling.

Can players crack the image recognition systems of well-known manufacturers and let AI make wrong decisions? If the players on the field can deceive AI, so can hackers, which will cause incalculable losses to manufacturers and users. The potential security hazards brought by the development of AI and cloud computing have caused thinking. Participants in the GeekPwn competition are geeks known as white hat hackers. They are highly skilled and will remind the company to fix when they find loopholes.

Geekpwn's Cloud Security Challenge is the world's first cloud security offensive and defensive competition based on the restoration of real cloud environments. At the competition site, Tencent Security Yunding Lab built a real full-stack cloud environment that can fully work and reproduce the mainstream cloud. The platform's architecture, technology, and system software and hardware environment preview the offense and defense on the cloud.

"Every app today, including AI, autonomous driving, and 5G communications, operates on the cloud, and the underlying is the silent support of the cloud. In this case, the cloud will be our very important infrastructure in the future. Infrastructure security Involving all aspects of future social operation, production and life, cloud security will be very important. "Li Bin, general manager of Tencent Security Yunding Lab pointed out.

Jinyun Security Competition is a competition that indicates the attack path for the players. Geeks may be school students or Internet company security teams. Unlike other events, cloud security and AI project teams rarely have personnel from enterprises, and are more composed of students at school.

For Wang Qi, there is a fault between traditional geeks and cloud security and AI. "Why do we emphasize the full attack path? Because people who know hackers have never touched the cloud environment. I will tell you the attack path, from this door to that door, but each door has a lock, you use the old unlocking method to open It. This process has helped the cloud computing industry, which is associated with traditional hackers again. "Wang Qi said in an interview with reporters.

Deep learning is an important factor in promoting the development of AI in the past two years. But the industry is already considering the security risks brought by the development of AI, and attacking deep learning models is also a popular item in geek competitions. It can be seen that some technical risks of AI are very prominent.

At the "CAAD CTF Image Confrontation Sample Offensive and Defensive Match" scene, multiple teams successfully used the image confrontation samples to launch attacks on the image classifier, causing the classifier to fail and tricking AI image recognition. All teams have broken through the YOLOv3 system, which is considered the "pinnacle of the target detection network." A team successfully deceived Clarifai's face recognition system with pictures, allowing AI to recognize Huang Jianxiang as Ivanka.

In awesome perspective, the significance of attacking deep models lies in prompting people to think more deeply about the real differences between machine and human vision. On the other hand, the unexplainable flaws of deep models themselves can help people understand the deep learning methods. limitation. Adversarial samples raise questions about deep learning and also provide an opportunity to modify deep models. This has profound implications for the military, safety and autonomous driving fields.

In 2020, Cisco also reformed the certification according to the progress of the Internet. PASSHOT also updated the latest certification in time. We can help you pass the latest CCNA 200-301 exam and CCIE Written exam.