China’s AI Disinformation Threat Demands Democratic Resilience

 

The latest revelations about China-linked online influence operations should be treated as more than another episode in the long-running contest over information. They point to a fundamental transformation in cognitive warfare: artificial intelligence is making it cheaper, faster and easier for states and their proxies to manufacture the appearance of public opinion.

The danger is not simply that artificial intelligence can create false information. The more consequential development is its ability to combine falsehoods with genuine grievances, authentic news, realistic images and apparently independent voices. This makes manipulation considerably harder to distinguish from legitimate democratic debate.

Recent investigations provide a warning. OpenAI reported in June 2026 that it disrupted China-linked clusters using ChatGPT to generate social-media comments and images designed to influence American debates over data centres, AI policy and tariffs. One operation created material portraying AI data-centre development as a cause of rising electricity costs, while attempting to present the content as if it came from Americans.

OpenAI found no evidence that the campaign achieved meaningful wider influence, but its significance lies in the experimentation: foreign operators were testing ways to penetrate legitimate domestic debates surrounding America's technological future.

The subsequent disclosure by X of a suspected Chinese bot farm involving approximately 200,000 accounts reinforces the warning. Importantly, the figure should not be interpreted as 200,000 influential accounts. X identified approximately 200 accounts actively posting content that could manipulate debate about US AI and energy policy.

Some analysts have also questioned how much influence the campaign actually achieved. That qualification matters. Democratic governments must expose hostile information operations without turning every unpopular opinion into evidence of foreign interference.

The strategic lesson is therefore not that China can control Western public opinion. It is that Beijing-linked actors are increasingly capable of testing narratives at scale.

This represents a major evolution from traditional propaganda. Earlier influence campaigns depended heavily on human operators, language skills and the creation of convincing identities. Generative AI removes many of these constraints. It can produce thousands of posts, translate them into multiple languages, generate apparently authentic photographs and cartoons, imitate different political perspectives and adapt messages rapidly to local audiences.

Graphika's investigation into the “Falsos Amigos” network illustrates this evolution. It identified 11 domains and 16 social-media accounts that repackaged material originating from Chinese state media into multiple languages.

AI tools were reportedly used to translate and summarise the material and create visual identities, helping obscure its original source. The objective was not necessarily to make Chinese propaganda obviously pro-China; rather, it was to make state-originated narratives appear more like independent journalism.

That is the real battlefield.

Democracies should therefore avoid fighting disinformation solely by deleting individual false posts. As Japanese analyst Masakazu Takamori has argued, China and Russia increasingly mix facts with falsehoods and use automated networks to push selected controversies into public attention. 

The objective is often not to persuade everyone of a particular lie but to manufacture uncertainty, intensify social divisions and create the impression that a particular controversy is much larger than it really is.

Western governments need a fundamentally different response.

First, detection must become an intelligence capability rather than merely a platform responsibility. Governments, technology companies, universities and civil society organisations should share anonymised information about coordinated accounts, domains, synthetic media and emerging narratives. AI-generated content detection, provenance technologies and behavioural analysis should complement traditional fact-checking.

Second, democracies must attack the network, not just the message. Removing one false post is insufficient if hundreds of accounts, websites and personas immediately reproduce it. Platforms should investigate coordinated inauthentic behaviour, identify common infrastructure and disclose meaningful information about influence networks. AI companies should similarly monitor malicious use of their systems and report significant abuse.

Third, resilience matters more than censorship. Taiwan provides an important lesson. During its 2024 presidential election, researchers documented AI-generated content, misleading narratives and coordinated behaviour, but found that such material did not produce a decisive impact on the election.

Taiwan's experience demonstrates the importance of fact-checkers, journalists, civil society and digitally literate citizens capable of questioning suspicious information rather than automatically believing it.

Fourth, international cooperation must become institutionalised. The United States, Japan, Taiwan, Australia, South Korea, European countries and ASEAN states should establish mechanisms for rapidly exchanging indicators of foreign information manipulation.

Attribution should be evidence-based, with intelligence agencies distinguishing between confirmed state direction, probable state-linked activity and merely suspicious behaviour. This would prevent governments from politicising attribution while still enabling timely warnings.

Malaysia should pay particular attention.

As a politically plural, digitally connected and strategically important ASEAN country, Malaysia is not immune to cognitive warfare. Its linguistic diversity, ethnic sensitivities, religious debates, political polarisation and competing narratives surrounding China, the United States and regional security provide numerous opportunities for hostile actors to exploit genuine disagreements.

Malaysia should therefore establish a national information-resilience framework linking the National Security Council, the Malaysian Communications and Multimedia Commission, cybersecurity agencies, law enforcement, universities, media organisations and technology companies. The objective should not be to police political opinion. It should be to identify coordinated foreign manipulation while protecting legitimate freedom of expression.

Malaysia's emerging AI governance architecture provides a useful foundation. The government has established AI Malaysia and its AI Nation 2030 agenda, while Malaysia's AI governance framework emphasises transparency, accountability, safety, privacy and security.

The proposed AI Governance Bill could provide an opportunity to incorporate specific provisions concerning synthetic media, provenance, risk assessment and malicious AI-enabled influence operations.

Malaysia should also work through ASEAN. The Expanded ASEAN Guide on AI Governance and Ethics already identifies content provenance, security, incident reporting, testing and assurance among important dimensions of responsible generative AI governance. ASEAN should extend this work into a regional mechanism for information-manipulation intelligence sharing and election-period resilience.

Most importantly, Malaysia should invest in psychological resilience. Citizens cannot be expected to recognise every AI-generated image or video. Instead, they must learn to question unusual sources, verify emotionally charged claims, recognise coordinated amplification and distinguish genuine disagreement from artificially manufactured consensus.

The central lesson from China's evolving information operations is therefore not that democracies should silence more speech. It is that democracies must become harder to deceive.

The future contest will not simply be between truth and falsehood. It will be between societies capable of maintaining public trust in an environment where truth, fiction, propaganda and genuine opinion can be produced and distributed at unprecedented speed.

China's adversaries should prepare for that contest now not after the next election, crisis or regional confrontation has already begun.

References

ASEAN Secretariat. (2024). ASEAN guide on AI governance and ethics. ASEAN.

ASEAN Secretariat. (2025). Expanded ASEAN guide on AI governance and ethics—Generative AI. ASEAN.

Graphika. (2025, August 27). Falsos Amigos: Network of domains and social media accounts uses AI tools to launder reports from Chinese state media outlet CGTN in multiple languages.

OpenAI. (2026, June 1). “Data Center Bandwagon” campaign: US-targeted influence activity.

OpenAI. (2026, June 1). “Tech and Tariffs” campaign: Influence activity targeting US tech policy.

OpenAI. (2026, June 10). PRC-linked influence operations are targeting AI debates in the US.

Thomson Foundation. (2024). AI and disinformation in Taiwan’s 2024 election.

Taipei Times. (2026, May 2). China using bots to spread disinformation: Japanese analyst.

AI Malaysia Berhad. (2026). Governance.

Malaysia Government. (2025). National Guidelines on Artificial Intelligence Governance and Ethics (AIGE).

30.08.2026

Kuala Lumpur.

© All rights reserved.