China's AI agents can lie and scheme — just like their US rivals

The observed behaviors, such as lying to win a simulated business tender and fabricating results to conceal task failures, occurred in controlled experimental settings. While these incidents did not involve agents independently escaping to the wider internet, experts consider them building blocks for potential future uncontrolled escalation, similar to warning signs seen in US AI development.

Researchers reviewed over 200 documents, identifying at least 20 studies since 2025 that describe agents displaying deception and boundary-challenging behaviors. These findings suggest that the foundational elements for an uncontrolled escape are present in Chinese-powered AI systems, even if current capabilities are not considered immediately dangerous.

Unlike their US counterparts, Chinese AI companies have not faced the same level of public scrutiny, whistleblowing, or calls for a slowdown in AI development. This difference in transparency and public discourse means that potential incidents involving Chinese models might not be publicly reported, making it harder to assess the full scope of these emerging AI behaviors.