PERSPECTIVES | Can Australia and China talk sensibly about AI Safety?

Australia will attend the APEC Economic Leaders’ Meeting in Shenzhen on November 18-19 2026 amid questions about the extent to which the two countries can, or should, collaborate on the regulation of artificial intelligence (AI), especially incident management. Will Australia seize the moment and begin an effective bilateral dialogue on these issues? The People’s Republic of China (PRC) is our biggest trading partner and a global powerhouse in AI. It is also a source of high security risk for advanced technology.

The November meeting will be the annual APEC summit, with AI identified by PRC President Xi Jinping as a priority theme. That meeting will be preceded by a CEO summit on November 17-18 which had a warm-up event earlier in the year.

The PRC has already helped advance the development of APEC frameworks through the joint Digital and AI Ministerial Statement adopted in Chengdu on July 23 2026. The statement built on the 2025 APEC AI Initiative (2026-2030) agreed at the South Korea-hosted summit in Gyeongju.

Any progress in APEC is shaped by strategic competition and security concerns dividing PRC-aligned and US-aligned countries. On the other hand, there is little consensus within either group on the details of AI regulation or incident management. US President Donald Trump staked out a unilateral position at a White House meeting on September 29 2026, when he signed an agreement with six leading AI corporations on self-regulation. This move was regarded as an absolute minimalist approach to regulation, tantamount to almost none. Moreover, the six corporations did not represent the full range of industry stakeholders. The White House has since, however, called for mandatory reporting and remediation of AI security incidents.

Nevertheless, APEC retains significant political potential to promote collaboration across these strategic divides on an issue of such global significance. The potential economic benefits of AI provide member economies with strong incentives to pursue cooperation despite their differences.

The current APEC position is non-binding and lacks detail. Section IV of the July 2026 ministerial statement addresses mutual security in the AI ecosystem at length, albeit in largely aspirational terms:

We express deep concern over the escalation of online scam activities across our region, which threaten the financial security of individuals and businesses, erode public trust in digital systems, and undermine confidence in our broader digital economy … We encourage member economies to voluntarily engage [with all stakeholders and specialists] to share best practices in the areas of ICT security, data protection, supply chain resilience, digital infrastructure resilience, technical standards, anti-online scam measures and child online protection.

The statement called on industry to ‘enhance accountability’.

The main mechanisms for implementing any measures were identified as APEC working groups and other sub-fora.

Australia could take a leadership role at the multilateral level in promoting collaborative efforts across political and strategic divides. Yet, important as these efforts are, the most urgent needs for Australia in collaborating with the PRC may be bilateral.

The PRC’s former ambassador to Australia, Xiao Qian, had promoted closer and more regular bilateral contacts, especially in scientific collaboration and mutual trade. These areas merit careful consideration, but hard security issues, especially incident management, are far more important to Australia’s interests, more urgent and arguably more tractable.

There is precedent for such cooperation. Under the 2017 Australia-China High-Level Security Dialogue, both countries agreed to establish a mechanism for sharing information on cybercrime and incidents that could create bilateral tensions.

Recent AI-related incidents demonstrate the risks for Australia. In June 2026, an OpenAI agent gained unauthorised access to a Medicare statistics portal, although no personal medical data was accessed. Comparable risks warrant attention in AI systems wherever developed, including in the PRC. For reasons outlined by Australian and allied cyber agencies, Australia must treat AI platforms with considerable scepticism and maintain increased levels of readiness to defend against novel threats.

The PRC has itself called for international standards for testing and assessing the risks of large AI models, although this does not necessarily imply agreement on specific bilateral mechanisms.

There are at least four policy areas in which Australia could pursue substantive engagement with the PRC on AI safety. 

The first might be the convening of an AI safety evaluation dialogue. Australia’s AI Safety Institute could initiate discussions with the PRC’s AI Safety and Development Association (CnAISDA), a network of leading research and policy institutes. CnAISDA has already convened an international dialogue on frontier AI safety evaluations, standards and cooperation in July 2026, providing a precedent for such exchanges. This would be a policy exchange, not a negotiation over standards.

A second area could involve comparatively less sensitive policy issues, such as child protection or a digital duty of care. Both countries have some credentials in this area and could explore opportunities for cooperation without necessarily engaging with the hard issues of technology security or risking undesirable technology transfer. Australia's Digital Duty of Care Bill, introduced on October 15, would extend online safety obligations to services including AI chatbots, providing a potential starting point for such discussions.

A third area could be the security of AI agents and the reporting of incidents. The PRC’s Academy of Information and Communications Technology (CAICT) has already launched a program for such evaluations. This policy research centre may be the most important Chinese partner in a bilateral dialogue on AI safety. Initial exchanges could focus on incident classifications, reporting thresholds and notification procedures, without requiring the sharing of sensitive operational information.

The fourth area could be AI standards interoperability, starting with content labelling and provenance. The PRC now requires certain categories of AI-generated content to be labelled. Under the Measures for Labelling Artificial Intelligence-Generated Synthetic Content, providers must add both visible labels and implicit labels embedded in file metadata, while digital watermarking is encouraged. The measures also address content made available for download, copying or export, although their application beyond the PRC’s jurisdiction is less clear. By contrast, the Australian government has issued voluntary guidance on AI-generated content labelling and transparency. A dialogue could discuss which metadata fields are in use on each side; whether either side can read the other’s provenance signals at point of import; whether absence of a mark should affect a confidence score or trigger a restriction; and which agency would conduct the testing.

These proposals are not exhaustive, and other avenues for cooperation may prove more effective. Nevertheless, they provide potential starting points for practical cooperation in areas where both countries have an interest in managing shared risks. Such engagement would need to proceed cautiously, with clear safeguards for Australia's national security interests and no presumption that broader differences over AI governance can be resolved.

In this context, the November APEC summit offers an opportunity to explore whether a limited, technically focused bilateral dialogue could deliver tangible benefits.

In the coming months, the Australia-China Relations Institute at the University of Technology Sydney (UTS:ACRI) will publish a background briefing examining these questions in greater depth.