Chinese AI Models Still Allowed to Run on Australian Government Systems

FILE PHOTO: The Deepseek logo is seen in this illustration taken on January 29, 2025. REUTERS/Dado Ruvic/Illustration/File Photo

Home Affairs officials could not rule out the possibility of government agencies using Chinese artificial intelligence (AI) models during the Joint Select Committee on Artificial Intelligence on Oct. 30.

Independent “Teal” Curtin MP Kate Chaney asked whether Chinese “open-weight” models could be used by government agencies or critical infrastructure providers right now.

Open-weight models are AI models that anyone can download and run on their own computers, rather than using them through the developer’s website or app.

Timothy Neal, acting first assistant secretary for protective security, said rules had applied to Commonwealth agencies since the government’s DeepSeek directive in February last year.

The directive ordered agencies to block and remove products, apps and web services from the Chinese AI company DeepSeek from all government systems and devices, following a threat and risk assessment.

Neal described the rules in broad terms, saying agencies needed the skills to deploy such models safely, they had to use certified secure cloud environments where classified information was involved, and had to show Home Affairs and the Australian Signals Directorate—the government’s cyber intelligence agency—that they could do so.

Labor MP Tim Watts pressed further, naming Chinese models Kimi (by Moonshot AI) and Qwen (by tech giant Alibaba) and asking whether Home Affairs would discourage agencies or other users of critical infrastructure from running them on their own systems.

However, Neal did not rule it out and said that in certain circumstances a government agency might run a Chinese AI despite the DeepSeek directive.

That would need to occur inside a protected, certified environment run by a major cloud provider, and there would have to be “suitable mitigations … in place.”

The security implications of using Chinese AI models have also been examined in the United States.

The findings raised questions about whether such models are suitable for use in sensitive software-development and government environments.

For the wider economy, First Assistant Secretary Greta Doherty said the department was still working out what safe use of open-weight models should look like for critical infrastructure and business.

“What does safe adoption of open weight models look like? What kind of mitigations might be needed to adopt that technology safely?” she said, adding that the models offered “significant benefits” for productivity and cyber defence.

Watts argued that open-weight models were increasingly being used by cyber defenders, and that access to a variety of models could make Australia more resilient at a time of rapid change.

He also asked how Australia would cope if a major AI model were withdrawn from the country, pointing to the period in June when Anthropic suspended access to two of its models to comply with U.S. export controls.

Doherty said she could not go into detail about a review of the government’s crisis plans, which is being led by the Prime Minister’s Office and Department.

Neal said the government signed its first “digital resiliency agreement” with Microsoft this year, setting out how services would keep running during a crisis, and was working on similar agreements with other companies.

We had a problem loading this article. Please enable javascript or use a different browser. If the issue persists, please visit our help center.

Source link