AI Is Lowering the Barriers to China’s Information Warfare Against Japan

When Japan’s Ministry of Defense released its 2026 Defense of Japan white paper, the document’s focus on China’s rapid advances in artificial intelligence drew predictable attention to the military domain. Missile guidance systems, autonomous platforms, and battlefield decision-making algorithms are all areas of legitimate strategic concern for Tokyo. Yet the most immediate danger to Japanese society may not come from AI’s application to conventional weapons systems at all. Rather, generative AI is quietly dismantling one of the last remaining obstacles to effective Chinese information operations targeting the Japanese public: the linguistic and cultural barrier.

The shift has been underway for some time, but it accelerated noticeably in late 2025. When Prime Minister Takaichi Sanae made comments in the Diet linking a Taiwan contingency to Japan’s own security, she provided a focal point for influence operators seeking to exploit Japanese political discourse. Since then, disinformation targeting Japanese political figures and domestic issues has grown—not only in volume but also in sophistication. Research by organizations including the Sasakawa Peace Foundation and OpenAI suggests that China-linked influence operators are increasingly incorporating AI tools into their workflows to adapt their activities to Japan’s online environment and improve operational efficiency.

The Erosion of the Linguistic Barrier

For years, one of the most reliable ways to identify Chinese-linked disinformation in Japan was simply to read it. Content often featured grammatical mistakes, awkward translations, and unnatural expressions that betrayed its foreign origin. Japanese readers could spot these telltale signs at a glance, and platform moderators could use them as heuristic filters. That era is coming to an end. Large language models can now correct grammar, adjust tone, generate culturally appropriate imagery, and produce multiple versions of the same narrative for different audiences within minutes.

The practical consequences of this shift are significant. What once required a team of translators, graphic designers, social media operators, and individuals familiar with Japanese political and cultural contexts can now be accomplished by a small number of operators using commercially available AI tools. The cost structure of cross-language information operations has collapsed, and the speed of production has increased correspondingly. If these capabilities are deployed to interfere with elections, discredit politicians, or amplify social divisions, their effects could extend far beyond any single piece of false information.

Documented Campaigns Against Japanese Political Figures

The evidence of these evolving tactics is already visible in reporting from Japan’s major media outlets. A special report by Nikkei in February found that more than 400 China-linked social media accounts spread negative content about Takaichi during Japan’s House of Representatives election. Separately, the Yomiuri Shimbun, citing research by Japan Nexus Intelligence, reported that under a different set of criteria, thousands of coordinated accounts on X were involved in sharing and amplifying related content. The narratives did not focus solely on attacking Takaichi personally. They also linked Japan’s efforts to strengthen its defense capabilities with concerns over social welfare, seeking to promote the argument that military expansion was coming at the expense of ordinary people.

This narrative framing is notable for its sophistication. Rather than relying on crude falsehoods that could be easily debunked, the campaigns sought to exploit genuine policy tensions within Japanese society. By connecting defense spending to welfare trade-offs, influence operators aimed to mobilize existing anxieties rather than manufacture entirely new ones. This is a hallmark of effective information warfare: it amplifies and distorts real social cleavages rather than inventing them from scratch.

The PAPERWALL Network and the Localization of Influence Operations

The Sankei Shimbun has continued to track PAPERWALL, a China-linked information network that includes 15 websites masquerading as Japanese news outlets. The continued operation of these sites, and the broader ecosystem surrounding them, points toward an increasingly localized approach to influence operations. Improvements in language, imagery, and the appearance of fake news sites are making it more difficult for ordinary readers to distinguish legitimate sources from fabricated ones. In the past, such accounts often left obvious clues. Some had been created only recently. Their Japanese reflected Chinese word order and linguistic habits, while others mixed simplified Chinese characters with Japanese usage. AI-generated images also frequently contained visual errors that made them relatively easy to identify.

Those weaknesses, however, are becoming less pronounced. The visual artifacts that once plagued AI-generated imagery have been substantially reduced in newer models. The grammatical errors that betrayed non-native authorship are being corrected by language models trained on vast corpora of Japanese text. The result is an information environment in which the traditional markers of foreign influence operations are fading, and ordinary readers are losing the ability to distinguish legitimate sources from fabricated ones.

A report released by OpenAI in February offered another important clue. The company found that China-linked actors had used ChatGPT to assist in planning an online influence campaign targeting Takaichi. Influence actors appear willing to incorporate different AI systems into their existing workflows whenever those tools make their operations more effective. What matters is not the nationality of the model. What matters is that generative AI is reducing the cost of cross-language information operations across the board.

The Taiwan Precedent and Societal Resilience

How, then, should Japan respond? Taiwan offers one useful point of reference. Taiwan has faced Chinese information operations for years and shares a Chinese-language environment with the mainland. Beijing therefore does not face the same linguistic barrier when targeting Taiwanese society that it encounters in Japan. Precisely for that reason, Taiwan was forced to develop social mechanisms for dealing with large-scale disinformation campaigns and networks of suspicious accounts.

Taiwan’s experience does not rely solely on government agencies or third-party fact-checkers. Individual users also identify, report, and block suspicious accounts, while civil society organizations and researchers continuously monitor information networks and patterns of coordinated behavior. This distributed model of resilience has proven more adaptable than centralized fact-checking alone, because it does not wait for false information to appear and then react. Instead, it builds the capacity to recognize coordinated influence operations before they gain traction.

Language and culture themselves can also serve as tools for identifying manipulation. Although Taiwan and China both use Chinese, there are clear differences in vocabulary, grammar, social experience, and political context. Chinese influence operators can use AI to reduce these differences, but they cannot easily reproduce the cultural background and lived experience accumulated within another society. Japan retains the same kind of advantage. Japanese language and social culture are highly contextual. Even if Chinese operators use AI to improve translation and content generation, that does not mean they can fully understand the linguistic and political differences between generations, regions, and social groups in Japan.

Toward Institutionalized Japan-Taiwan Information Security Cooperation

This is also why information-security cooperation between Japan and Taiwan should become more institutionalized. The two sides could deepen exchanges on identifying coordinated accounts, detecting deepfakes, analyzing foreign influence networks, and monitoring information environments during major disasters, elections, or political crises. The threats they face are increasingly similar, and the lessons learned in one context are often directly transferable to the other.

For South Korea, the implications of this analysis are direct and sobering. Seoul faces analogous China-linked influence operations targeting its own society and elections, and the same AI-driven erosion of linguistic barriers applies to the Korean language. Korean information environments are if anything more exposed, given the intensity of domestic political polarization and the speed of the country’s digital media ecosystem. The Japanese experience—and the Taiwanese experience before it—offers a template for building societal resilience that Seoul would be wise to study closely. The specific incidents and numbers differ, but the strategic pattern is consistent across the region.

AI is making Chinese information operations cheaper, faster, and easier to deploy across language barriers. But it still cannot fully replicate the cultural knowledge and lived experience embedded in a society. Japan therefore does not need to build a wall capable of blocking every piece of false information. It needs to build an information environment in which foreign influence operations are less likely to succeed. As Japan and Taiwan confront increasingly similar information-security threats, sharing their experience in identifying and countering Chinese influence operations could become an increasingly important part of regional security cooperation.

This article was produced with AI-assisted research and editorial support. Sources: The Diplomat (Cheng-Pu Kao, "AI Is Lowering the Barriers to China's Information Warfare Against Japan"), Nikkei, Yomiuri Shimbun, Sankei Shimbun, OpenAI.