APEC open-source AI statement backs models with strong security

The U.S., China and the other members of APEC backed APEC open-source AI cooperation in a ministerial statement issued Thursday in Chengdu, China, while tying that support to stronger security controls. The 21 economies said artificial intelligence development should respect security, data protection and intellectual property rights, a formulation that points to tighter public oversight as more capable models are released for broad use.

The statement, named for the host city, said APEC members would support open-source AI models and projects that use strong security assurance in their development and deployment. That language gives governments room to endorse wider access to AI tools while still pressing for testing, transparency and operational controls.

Open-source AI models can be downloaded and used without the pay-to-access structure common among closed commercial systems. In practice, open access can speed adoption by developers, companies and governments, while also raising concerns about misuse, cybersecurity and the protection of proprietary technology.

What did APEC say about open-source AI?

APEC’s Chengdu statement endorsed cooperation around open-source AI and related digital infrastructure, while stressing that such systems should be developed with security and rights protections. Li Lecheng, China’s minister of industry and information technology, said the document was the first APEC artificial intelligence statement to include open-source cooperation at the ministerial level.

Li, who chaired Thursday’s meeting, said participants acknowledged AI-related risks and supported more discussion among governments, business and academia. He cited cybersecurity, supply chain resilience and online scams as areas where information-sharing would be encouraged.

The agreement arrives as both Washington and Beijing are reported to be seeking more control over domestic AI capabilities. China has been a leading force in recent open-source AI releases, while U.S. officials have alleged theft from American technology in connection with Chinese AI development.

Chinese systems including DeepSeek and GLM 5.2 are available to download and use, according to CNBC. By contrast, some U.S. developers, including Anthropic, offer closed AI models that users access through paid services rather than by downloading the model.

Why governments care about open models

Winston Ma, an adjunct professor of law at New York University, told CNBC the Chengdu statement showed explicit support for open-source ecosystems and computing infrastructure. He said the Asia-Pacific region is moving toward open-weight systems paired with state-coordinated energy, telecom and digital infrastructure.

Wei Sun, principal analyst for artificial intelligence at Counterpoint Research, said getting all 21 APEC economies, including the U.S. and China, to support trusted open-source AI was meaningful. She said the wording on strong security assurance allowed more cautious governments to back open models while still requiring testing, transparency, data protection and controls over deployment.

Sun said the policy debate is shifting from whether AI should be open or closed toward which actors can build open systems trusted enough for public-sector and enterprise use. That distinction matters for companies and governments deciding whether to adopt AI tools whose underlying code or model weights are more widely accessible.

APEC business representatives also raised concerns about the next wave of computing risks. Janet De Silva, chair of the APEC Business Advisory Council’s digital and innovation working group, urged ministers to prepare for the quantum computing era and said trusted encryption was needed to protect the financial system while supporting competitiveness.

This story draws on original reporting from CNBC.