AI "Crossing the Line"? The International Community Explores Setting Up "Guardrails"

Author:Qiao GAO Release date:2026-11-17 17:14:14Source:FDDI

On October 13, 2026, Qiao GAO, a reporter from the People's Daily Overseas Edition, interviewed Tianjiao JIANG, Researcher at the Center for Global AI Innovative Governance (CGAIG) and Associate Professor at Fudan Development Institute (FDDI), asking him to analyze issues such as AI security risks and international cooperative governance. JIANG believes that the development of AI foundation models has entered a new stage. Multiple incidents of AI crossing the line show that the model out of control risk, originally belonging to sci-fi imagination, is becoming a realistic challenge. Traditional security designs for single models are struggling to cope with hidden dangers brought by multi-agent collaboration, as multiple interacting agents may bypass safety guardrails. Meanwhile, there is also the problem of unclear liability definition after AI goes out of control. AI risks are mainly divided into four categories: abuse, loss of control, social, and geopolitical and military risks. Currently, global AI governance presents fragmented characteristics. Although multilateral platforms like the UN and G20 have relevant initiatives, a unified coordination framework has not yet been formed, with significant differences in regulatory paths among countries. He suggests that in the face of rapid AI development, security supervision and international cooperation are urgent. Development and security must be given equal weight. We should promote the establishment of a globally recognized third-party independent evaluation and mutual recognition mechanism for frontier models, an international security incident reporting and disclosure mechanism, and perfect AI classification and grading standards. Stepping out of the mindset of judging risks purely by parameter size, we should focus on substantive risk features such as autonomous iteration, autonomous network attacks, and high-risk coordination among multiple agents to draw safety red lines for AI and jointly set up guardrails for AI through international collaboration.


Translated by FDDI

Full text in Chinese available at:

https://fddi.fudan.edu.cn/30/10/c18965a798736/page.htm