# Bilateral AI Partnerships Alone Won't Secure Australia. Here’s a Multilateral Alternative - RAND

*Источник: RAND*
*Дата: 2026-07-14*
*Язык: en*

**Кратко:** Released in December 2025, Australia's National AI Plan (PDF) outlines an ambitious vision: harness the opportunities of AI, spread its benefits nationwide, and keep Australia safe. A theme throughout the plan is the push for deeper cross-border and cross-sector collaboration, reflecting Australia's familiar ambition to be a ‘partner of choice.'
Beyond the few bilateral commitments and pledges to norms and safety, what could meaningful regional AI collaboration look like?

Released in December 2025, Australia's National AI Plan (PDF) outlines an ambitious vision: harness the opportunities of AI, spread its benefits nationwide, and keep Australia safe. A theme throughout the plan is the push for deeper cross-border and cross-sector collaboration, reflecting Australia's familiar ambition to be a ‘partner of choice.'
Beyond the few bilateral commitments and pledges to norms and safety, what could meaningful regional AI collaboration look like? The answer lies in federated learning, an AI model that allows cross-border collaboration on transnational security issues whilst preserving data sovereignty. Becoming an early proponent of a cross-border federated learning model could position Australia advantageously as a key partner for AI-enhanced collective security.
While there are many possibilities, such as cloud infrastructure, more regulatory coordination, and joint data centers, federated learning is underexplored but currently provides promise. Federated learning is a machine learning technique in which trusted partners can collaborate using a decentralized approach to data, training a shared AI model, to address cross-border security risks, such as on cyber incidents, mis- and disinformation, financial security, or open and secure maritime waterways, without sharing potentially sensitive raw data.
Federated learning, proposed in 2017 by McMahan (PDF), was originally designed to train on decentralized mobile data, and has since proven viable in sensitive domains. First pilots in medical diagnostics and fraud detection demonstrated federated learning's technical viability and accuracy. Australia has an increasing number of research groups and projects using federated learning in health and finance. Early successes show the potential of federated learning in threat detection and highlight its role in addressing critical privacy and security issues. For instance, to counter scammers and safeguard the assets of the public, Taiwan's Fubon Financial Holdings initiated an ‘Eagle Eye Fraud Prevention Prediction Model,' utilizing AI fraud detection technology through federated learning. Its model achieved an 80 percent success rate in detecting fraud, preventing NT$360 million reaching malicious groups. Research also shows federated learning can detect deepfakes and disinformation, and theoretical proposals suggest it strengthens cybersecurity against threats (PDF). Yet, a cross-border federated learning approach to counter security threats remains underutilized in Asia and the Pacific.
Early successes show the potential of federated learning in threat detection and highlight its role in addressing critical privacy and security issues.
This gap has direct implications for Australia's AI ambitions. Australia's National AI Plan aims to maximize AI's benefits while reducing its potential harms to society. The plan presently appears to emphasize bilateral collaboration to improve local AI capabilities and position Australia as a trusted leader, with efforts focused on multilateral norms and governance. Tangible technical cross-border collaborations appear largely mostly bilateral, like the strategic partnerships with the UK, South Korea (PDF), and India (PDF) that focus on innovation and research. But while the bilateral initiatives are valuable, Australia should also pursue tangible multilateral projects that actively counter key risks amplified by AI. Australia needs multilateral AI projects that pool expertise, data, and infrastructure to counter key threats, while preserving strategic autonomy from dominant tech providers, whether commercial monopolies or state-backed competitors.
The case for multilateral technical collaboration is straightforward: The threats Australia faces transcend bilateral frameworks. Moreover, the capacity and incentives to collaborate already exist in the region. Key regional threats, such as deepfakes, fabricated media, and disinformation, are increasingly enabled and amplified through AI. The Asia and the Pacific region faces frequent cyberattacks, now driven by adaptive AI malware that automates the entire attack lifecycle. AI in the maritime domain asymmetrically benefits advanced countries, enhancing their ability to monitor, track, and interpret maritime activities.
These shared threats unfold amid US-China AI competition, risking the strategic autonomy of middle powers like Australia. Fortunately, countries in the Indo-Pacific are highly capable and have invested significantly into sovereign AI. Singapore and South Korea lead global AI readiness worldwide, while Australia leads in the Pacific. Furthermore, Singapore leads in the AI governance space. Taiwan is a key semiconductor producer; South Korea hosts global AI champions, and Japan possesses extensive competitive computing power (PDF). With complementary strengths, shared goals, and similar threats, pursuing joint technological cooperation is a natural choice, and becomes especially compelling when it can be done without compromising sovereign interests.
The federated learning model allows for multilateral engagement on AI collaboration without compromising data sovereignty. Its key feature of decentralization resolves regulatory, privacy, and strategic concerns that arise when countries restrict raw data being sent abroad. Federated learning localizes, aggregates, and anonymizes data, enabling compliant cross-border collaboration while easing integration. This machine learning technique also reduces communication costs versus related distributed training methods, which is particularly relevant when training large datasets common in multinational collaborations. Federated learning's decentralized design lowers data transfer, storage, and energy costs, reducing expenses and carbon footprint. Amid increased scrutiny over AI's vast water consumption, this advantage addresses sustainability policy concerns.
The federated learning model allows for multilateral engagement on AI collaboration without compromising data sovereignty.
In theory, federated learning enables cross-border collaboration on sensitive data—such as data related to disinformation, maritime tracking, or cyber incident monitoring—while preserving strategic autonomy and data sovereignty. Australia should seriously consider this approach, as federated learning can securely leverage diverse cross-border data to improve outcomes while respecting strict privacy laws. This enhances the aggregated model's ability to detect threats in real time in ways that would be unachievable unilaterally. As shown in the National AI Plan, Australia has signed numerous ethical and normative AI agreements, which could become tangible by implementing low-risk projects, translating intentions into meaningful action. It would also foster knowledge exchanges, accountability, and trust-building, creating a basis for broader cooperation.
Participation in this initiative would boost Australia's resilience against destabilizing threats and foreign interference. Australia's extensive data-gathering abilities, especially compared with smaller Pasifika partners, allows Canberra to offer unique data to a federated learning model, that could result in a wider benefit for the whole region. Early adoption of a cross-border federated learning model could make Canberra a ‘partner of choice' in the Indo-Pacific, encouraging Pacific nations to follow. While Pacific Island nations lack the AI infrastructure for near-term participation, Australia's contribution of Pacific-relevant threat data, including data on maritime surveillance patterns and regional disinformation campaigns, indirectly strengthens their security through the collective model. Acknowledging the interconnectedness between Australia and the South Pacific, a multilateral federated learning model would thus enhance both individual and regional security across Australia and the South Pacific.
Federated learning supports strategic autonomy, letting Australia share threat intelligence while retaining data sovereignty—made possible without relying on tech giants or symbolic bilateral frameworks. It also offers collective resilience, improving domestic detection and providing insights into partners' threat landscapes amid shared AI-driven risks. This creates a shared shield, allowing participants to leverage each other's threat detection and implement preventive measures. Australia and its partners share a symbiotic relationship, where mutual safety strengthens overall regional security.
Federated learning has the potential to signal a strategic and normative shift in how middle powers collaboratively contribute to the AI ecosystem. If Australia wants to strengthen its AI sector, as envisioned by the National AI Plan, it must invest in collaborations that support key strategic interests. Federated learning could be a suitable way forward.

[Оригинал](https://www.rand.org/pubs/commentary/2026/07/bilateral-ai-partnerships-alone-wont-secure-australia.html)