Profile: Shanghai AI Lab: Driving both AI safety and development

China was notably absent when AI safety institutes (AISIs) from nine countries and the EU met for the first time in November 2024 as a network to collaborate on AI safety research, testing and guidance. Experts are speculating whether China will designate an institution to coordinate its AI safety work or create a nationwide network of AISI-like entities. Regardless of the path China takes, the Shanghai Artificial Intelligence Laboratory (上海人工智能实验室, SHLAB) is undoubtedly an important player.a relevant player.  

Headed since July by former chief scientist of IBM’s Watson Group and later director of AI research at JD.com, Zhou Bowen (周伯文), SHLAB is one of China’s New R&D Institutes. While Big Tech drives AI development in the US, in China, state-backed research labs like SHLAB are frontrunners in advanced AI development – and increasingly in governance. They attract world-leading talent and can count on excellent resources and infrastructure.

Zhou is both a proponent of Artificial General Intelligence (AGI) and an advocate of strong safety guardrails around frontier model development. In his inaugural speech at the 2024 World AI Conference, he urged a rebalancing of AI investment toward safety research and technical measures. He compared AGI to nuclear fusion and argued that safety is a global public good.  

SHLAB is spearheading technical research on AI safety, according to Concordia AI. For example, they have published benchmarks to evaluate large language models in terms of human value alignment, robustness to adversarial attacks and safety from risks ranging from generating toxic content to malicious use in weapons of mass destruction. SHLAB is also leading LLM-relevant standards work in China.  

Although the EU participates in the AISI network, the European AI Office lacks the depth of technical expertise at institutions like SHLAB. Engaging with SHLAB and similar organizations in China would be in the interest of Europe’s own ecosystem development and governance impact.