Open-source AI has been absent from most multilateral security frameworks because model weights, once published, cannot be revoked or tracked, making traditional enforcement architectures inapplicable. That gap narrowed at APEC, where the U.S. and other member nations backed a statement endorsing open-source AI cooperation with "strong security" conditions attached. Li Lecheng, China's minister of industry and information technology, said it was the first APEC statement to address open-source cooperation at a minister level.
The constraint the statement is working around
When model weights are freely downloadable, access management becomes theoretical. Any party can download and redeploy the same base model without logging activity against the original developer. Multilateral security frameworks are built on the premise that access to sensitive technology can be tracked and, if necessary, cut off. Open-source AI breaks that premise at the architectural level. The "strong security" qualifier is the statement's attempt to hold both positions: endorse open development while preserving a policy hook for member governments to impose downstream controls.
What the minister-level classification adds
Li Lecheng described the APEC statement as the first at minister level to include open-source cooperation language. China hosted the summit. Prior APEC discussions had not produced comparable text at that tier, according to Li Lecheng. Ministerial statements at APEC are consensus documents, not self-executing treaties, so individual member governments retain authority to define what "strong security" means inside their own regulatory frameworks. When the U.S. and China both appear on the same side of an open-source AI formulation at a senior policy level, that text becomes a baseline. Subsequent discussions either extend it or break from it explicitly, rather than starting from silence.