Safety

China and US Urged to Cooperate on AI Safety Amid Cyber Risks

A major AI conference in Beijing highlighted the need for global collaboration to manage risks from advanced models, with experts warning of potential chaos if nations fail to work together.

Close-up of a blue screen error shown on a data center control terminal.

Photo: panumas nikhomkhai / Pexels

At a recent AI conference in Zhongguancun, Beijing, experts emphasized the need for the US and China to collaborate on AI safety, given the increasing risks posed by advanced models. The event, organized by the Beijing Academy of Artificial Intelligence, featured discussions on cybersecurity, agentic models, and the potential for AI to cause widespread harm if developed without oversight. Stephen Casper, a computer scientist at MIT, warned that AI's global impact means cooperation is essential to prevent catastrophic outcomes. 'AI doesn't need a Chernobyl moment,' Casper said, stressing the importance of shared safety standards. The conference underscored the urgency of addressing systemic risks as AI becomes more integrated into daily life and more powerful.

The US has historically viewed China's AI advancements as a national security threat, leading to restrictions on chip exports and access to advanced models like Mythos and Fable 5. Anthropic revoked access to these models after the US government ordered them to block foreign nationals. Meanwhile, Chinese companies like Z.ai and Alibaba have led in developing open-weight AI models, such as GLM 5.2, which include agentic and coding capabilities. A cybersecurity firm in China recently claimed to have developed an AI model with hacking capabilities comparable to Mythos. These developments highlight the growing complexity of balancing innovation with security.

Lin Yun, a professor at Shanghai Jia Tong University, noted that while hackers may gain an advantage in the short term, new countermeasures, including AI-driven defenses, could shift the balance back toward security. Yun emphasized the importance of international cooperation, stating that shared understanding of risks can help establish common safety principles. 'The key is to find areas where sharing can reduce systemic risk without exposing sensitive operational details,' he said. As open-weight models become more powerful, ensuring they are secure, free of backdoors, and meet safety standards will be critical. The industry is now facing an inflection point where even less advanced models could pose significant dangers if not properly regulated.

Source: wired

Key points

Stephen Casper, a computer scientist at MIT, emphasized the need for global cooperation on AI safety to prevent catastrophic outcomes.
The US government ordered Anthropic to restrict access to its most powerful models, Mythos and Fable 5, due to national security concerns.
Chinese companies like Z.ai and Alibaba have led in developing open-weight AI models, such as GLM 5.2, with agentic and coding capabilities.
A cybersecurity firm in China claimed to have developed an AI model with hacking capabilities comparable to Mythos.
Lin Yun, a professor at Shanghai Jia Tong University, noted that while hackers may gain an advantage, new countermeasures could shift the balance toward security.
The industry faces an inflection point where even less advanced open-weight models could pose significant dangers if not properly regulated.

Source: WIRED Read the original →

WRITTEN BY

Nadia Rahman

AI Safety, Alignment & Policy

Nadia follows AI safety, alignment, regulation, and the policy debates shaping the field.