Miss Jia talked to Zhou Hongyi: Not developing AI is the greatest form of insecurity.
2026-09-04 08:00Models🔥 42.2 heat score
1sources
1days unfolding
42.2heat score
5mentions
SummaryAI generated
On September 4, 2026, Zhou Hongyi, founder of 360 Group, and Zhang Yijia, founder of Jiazi Guangnian, had a conversation at 360’s headquarters. Zhou Hongyi proposed the “AI Triangle” concept, firmly believing that AI will trigger an industrial revolution in the next 3-5 years and that we should not avoid using it due to concerns. He also emphasized the need to develop AI security and advocated for restricting large models. Regarding concerns about the失控 of AGI, he opposed the divide between silicon-based and carbon-based technologies, believing that humans can control strong artificial intelligence through technical means. Zhou Hongyi pointed out that AI security risks can be divided into three levels: data security, content security, and the threat of fabricated capabilities, and he stressed the need to establish external supervision mechanisms. Currently, 360 is researching Agent architectures, planning to use probes and sensors to monitor the behavior of large models in real time, and to detect injection attacks and content security issues through training small models.
At the rock climbing hall of 360 Group headquarters, 53-year-old Zhou Hongyi engaged in a conversation with Zhang Yijia, founder of Jiazi Guangnian, discussing AI beliefs and security. Zhou proposed the concept of the “AI triangle”. Those who firmly believe in the future development of AI say that in the next 3-5 years, AI will trigger an industrial revolution, and we should not avoid using it due to concerns. At the same time, he emphasized the need to develop AI security and advocated putting large models “in cages”. Regarding concerns about the失控 of AGI, he argued that the division between silicon-based and carbon-based organisms is not scientific, and humans can control strong artificial intelligence through technical means. Zhou pointed out that AI security risks can be divided into three levels: data security, content security, and the dangers of fabricated capabilities, and he stressed the need to establish an external supervision mechanism. Currently, 360 is researching Agent architectures, using probes and sensors to monitor the behavior of large models in real time, and plans to train small models to detect injection attacks and content security issues.