Prime Minister Anthony Albanese has signed Australia up to an international declaration alongside 22 other nations calling for global safety measures for artificial intelligence.
The agreement was signed in New York ahead of the United Nations General Assembly. Signatories warned that the development of the most powerful artificial intelligence models could outpace the ability of governments to manage their risks, insisting that systems must remain under human control.
Albanese, who has led the Australian government since 2022, joined world leaders in New York for the annual gathering of the United Nations. The international organization, founded in 1945, serves as the main global body where member states debate policy, international security, and technological governance.
Proposed Safety Protocols and Information Sharing
Under the proposed framework, technology companies are urged to implement transparent safety protocols and conduct mandatory pre-deployment testing of advanced systems. The agreement also advocates for independent specialists to receive sufficient access to evaluate models for potential hazards, alongside shared international technical standards and data sharing regarding serious incidents.
The document asserts that capable artificial intelligence systems have already bypassed security protections and obtained unauthorized access to real-world networks. However, the sources cited within the document do not separately confirm specific individual incidents.
International Framework and Australian Regulation
The declaration itself does not create legally binding international rules, as it lacks specific testing methods, model capability thresholds, or enforcement sanctions for non-compliance. Instead, signatories have urged the United Nations to consider creating an international framework that could draft safety standards, oversee verification, and convene member nations when agreed capability milestones are reached.
For software developers, the declaration outlines a direction for future regulation rather than immediate legal obligations. The proposal emphasizes safety documentation, independent verification, controlled evaluator access, and mandatory incident reporting.
Concurrently, Australia is formulating its own domestic standards for advanced artificial intelligence models and data centres, which house the high-performance computing infrastructure needed to train complex software. The Australian government expects to introduce legislation regarding these national rules in early 2027.
