Artificial intelligence is a weapon of mass disinformation

Using AI to search for something on the internet is simple and fast, but if it’s no longer necessary to search for ourselves, it’s also—and above all—because we’re delegating to a machine the task of choosing the right source of information.

I have no major objection to using it to fill out an Excel spreadsheet, but more and more users rely on it to stay informed. And to be honest, it’s happened to me too.

According to this TF1 article45% of consumers trust AI to find information about a brand. The problem is that AI doesn’t reason like a human would: it’s trained on data meticulously selected by its owners and follows a “system prompt” (a kind of supreme instruction it’s not supposed to violate).

Take the example of DeepSeek, a Chinese model. It’s impossible to make it say that Taiwan is an independent country.

I tried to get more information on the topic by opening a new chat to ensure the previous response wouldn’t influence the next one:

This response bothers me. No sources are cited, and the AI simply recites Chinese government propaganda. DeepSeek can perform online searches, but it clearly chose not to for this response. In a world where China dominated the AI market, such responses would be the norm.

For comparison, here’s the same question asked to Mistral AI’s Le Chat:

full response here)

In the screenshot, Le Chat cites its source: the media outlet 20 Minutes. I use it daily and it tends to cite reputable, reliable media when possible. Admittedly, media outlets aren’t always free or independent—a whole other issue—but at least they offer multiple perspectives on the same topic.

It’s hard to rewrite history books, but training an AI model on biased data is within the reach of any major state power—or even a corporation. Incidentally, we can see that American artificial intelligences are highly popular among the French:

In such a situation, could we perhaps speak of ideological interference? If American services spread propaganda through their AI, we’d be subjected to it. But they’d never do that… right?