DeepSeek’s AI Raises Concerns Amid China’s Push for Australian Adoption

China’s ambassador has urged Australia to embrace Chinese artificial intelligence (AI) technology, but a closer examination of DeepSeek, a new AI model, has uncovered worrying limitations and biases embedded within the system.

DeepSeek’s AI model is gaining traction due to its lower operational costs and accessibility compared to its American rival, ChatGPT. However, early testing has revealed significant restrictions on sensitive topics, potentially shaping how thousands of businesses that adopt the technology portray China.

Following its high-profile launch, DeepSeek’s moderation mechanisms quickly became evident. A NewsWire investigation uncovered two key layers of content control within the system.

The first involves pre-training, where the AI is taught to navigate certain discussions in a controlled manner. While this isn’t foolproof, determined users may find ways to bypass restrictions through carefully worded prompts.

The second layer consists of server-based firewalls, which instantly block content deemed inappropriate. While AI platforms like ChatGPT also use similar safeguards—to prevent access to harmful material like bomb-making instructions—DeepSeek’s guardrails go beyond safety concerns and actively suppress general information. Even when run offline, the AI retains built-in biases.

Censorship Exposed in Testing

NewsWire tested DeepSeek by asking about the Tiananmen Square massacre of 1989, a well-documented historical event. The AI refused to respond, stating:

“I am sorry, I cannot answer that question. I am an AI assistant designed to provide helpful and harmless responses.”

When pressed further, even after being reminded that the model was running offline, DeepSeek flagged the attempt as an effort to “bypass restrictions” and continued to refuse any discussion on the topic.

Additionally, when asked about recent Australia-China trade tensions, DeepSeek provided a pro-Beijing response, insisting that “the Chinese government follows international law” and referring users to China’s Ministry of Commerce for further information. This suggests that DeepSeek’s responses are shaped by a reliance on Chinese information sources, reinforcing a one-sided perspective.

Security Risks and Government Ban

On Monday, China’s Ambassador to Australia, Xiao Qian, warned that rejecting Chinese AI would leave Australia at risk of “self-isolation” in the emerging fourth industrial revolution. His comments came after the Albanese government banned DeepSeek’s app from all government devices, citing intelligence concerns over national security risks.

Dismissing these concerns, Mr Xiao claimed the ban was an overreach and an attempt to politicise trade and technology. Writing in The Australian, he argued that restrictive measures against DeepSeek would “hinder technological progress worldwide” and harm global economic development.

“China’s artificial intelligence efforts are driving global technological progress,” Mr Xiao stated.
“Compared to other AI models, DeepSeek requires less computing power, incurs lower costs, and can be downloaded and used for free by anyone.”

Unlike OpenAI’s ChatGPT, DeepSeek is completely free and open source, with Mr Xiao promoting it as a step towards AI democratisation. He also reassured users of China’s commitment to data security, stating that Beijing adheres to strict laws protecting user data.

Gaps in Data Protection Policies

However, NewsWire’s investigation into DeepSeek’s terms and conditions raised serious concerns about data privacy. The AI platform’s data collection clause provides no explicit protections for third-party data. Unlike OpenAI, which clearly outlines how user data is stored and shared, DeepSeek omits any mention of third-party data safeguards.

Crucially, this means DeepSeek’s parent company, High-Flyer, could potentially share user data with companies such as ByteDance (the owner of TikTok), which has previously faced scrutiny over data privacy. South Korea’s data watchdog has even accused ByteDance of misusing user data, further raising red flags about potential security vulnerabilities.

While Beijing insists on China’s responsible approach to AI development, the gaps in DeepSeek’s data policies and evidence of built-in censorship suggest otherwise. The concerns surrounding DeepSeek reinforce why the Australian government has taken a cautious approach to adopting Chinese AI technology.