Why we can't rely on governments to handle AI safety

The latest security scare involving Chinese superapp WeChat highlights the inadequacy of governments to handle AI safety effectively. A potentially devastating bug in WeChat, known as a worm, was revealed by US security company Calif, which can hijack an account and spread through victims' contacts, regardless of user actions. This incident occurred weeks after the AI-powered hack of Hugging Face, which used a Chinese open-source model to determine the cause of the breach.

The episode emphasizes that cyber vulnerabilities do not align neatly with state borders, as stated by US-China tech expert Samm Sacks. Despite both AI superpowers, China and the US, having equal incentives to manage cyber risks, trust between the countries remains nearly nonexistent. Additionally, the definition of AI safety differs between Beijing and Washington, with the former seeking to define the discourse according to its own worldview.

Consequently, a more technical approach to AI safety, driven by private companies, researchers, and technologists, rather than ideology, can provide a stronger foundation for any bilateral agreement. In the absence of such an approach, the private sector will have to rely on unofficial cross-border communication and collaboration to prevent the next WeChat worm from escaping the lab.