Until 2028: China's plan to take control of the answers provided by the chat
People use AI to answer questions on a wide range of topics. But behind the answers received, there is a struggle between different countries over the content on which the models are based. To win the war of narratives, China is willing to invest capital, and so far it is working for them.
When you ask a chatbot about a territorial dispute, a historical event, or a political position, the answer received from the artificial intelligence may appear neutral, reasoned, and completely objective. But under the technological hood that provides the precise answer, a shadow war is being waged between various parties over every single word.
This is a well-funded and well-organized state effort: from investing resources to create models and build datasets, to direct pressure on technology companies. The goal is not just to influence what people see, but to determine what the machine knows.
Recruitment and data dissemination: China prepared for war
One of the most prominent examples of this is China, which is working to become a global data powerhouse for training artificial intelligence models. Beijing wants to compete with the Western narrative led by the USA and embed the values of the Communist Party into global AI systems.
As recently reported by The New York Times, China understands that the world's leading AI models are trained primarily on English-language data and reflect a Western worldview. From Beijing's perspective, this is a strategic vulnerability that causes Western narratives on issues of human rights, the status of Taiwan, and politics to prevail over the Chinese position.
According to the report, China has presented a plan to turn the country into a supplier of training data by 2028. The goal is to create high-quality databases in more than two dozen strategic fields, recruit experts for data labeling, open academic courses, and distribute open-source databases — all with information consistent with China's values.
Flooding with content: how it works
Osher Cohen, founder of the company Yo Secure and a cyber expert, explains: "Certain AI models, especially when connected to internet search, look for sources before they formulate an answer. If any party manages to generate a lot of content online around a certain narrative, ensure it looks credible, and has a high presence in search, it may increase the chance that this information will enter the process from which the AI assembles the answer."
Cohen explains that the real problem is transparency: "The real problem from the user's perspective is that it is difficult to see this whole process. We receive one organized and confident answer, but we don't always know why these specific sources were included. The question today is no longer just whether the AI is wrong, but who managed to influence the information that the AI saw before it answered us."
Fighting disinformation: the report on Israel
Politico reported that Israel launched an influence campaign in the USA to feed language models with positive information about the activities of the IDF and the war in Gaza, against the backdrop of declining public support. According to the report, the National Public Diplomacy Directorate employed a French advertising agency, Havas Media, with a subcontractor called Piro. 100,000 dollars were invested in the campaign, and a website for the "Hanover Institute for Public Policy" was established, containing dozens of anonymous articles in a Q&A format, which helps the absorption of content into models.
Recall that in 2024, OpenAI revealed that it had identified and blocked five covert influence operations originating from Russia, China, Iran, and Israel, which used its AI tools to streamline and accelerate propaganda activity.
Heavy concern ahead of the elections
The upcoming elections in Israel will be a case study regarding the ability of models to influence the public. Dr. Tehilla Shwartz Altshuler, a senior fellow at the Israel Democracy Institute, explains: "The level of chatbot penetration in Israel is among the highest in the world, while digital literacy is mediocre at best. People use the chat as an advisor to get information on issues that are on the agenda, including election matters. The fear is that this will become a significant influence tool, as the result given in a chatbot is personal, making it harder to see the fingerprints of an influence campaign in real-time."
