Covert Influence Operation

A Covert Influence Operation is an organized campaign intended to push the opinion of many people in a certain direction. What makes it distinctive is that whoever is behind it stays deliberately hidden. A post looks like the opinion of a private individual, but in truth it comes from a government, an intelligence service, a political party, or a company. The German term for this is verdeckte Einflussoperation. It’s important to distinguish this from ordinary advertising or propaganda: there, you know who is speaking. Here, concealing the source is itself the actual trick.

Why disguised senders are more dangerous than open ones

People judge a statement by who is making it. Most people take a claim from a government broadcaster with a grain of skepticism. The same claim from an apparently ordinary user seems more credible. It is precisely this mental shortcut that covert campaigns exploit. In a sense, they steal the trust we extend to real fellow human beings.

There’s a second effect on top of that. When hundreds of accounts voice the same opinion, it appears more widespread than it actually is. Experts call this an artificially manufactured majority, or astroturfing, after artificial turf that mimics real grass. Many people adjust their own position to match what they believe is the majority view. So a campaign doesn’t need to directly convince anyone. It’s enough to create the impression that almost everyone thinks this way.

Elections, wars, and social conflicts are especially affected. Often the goal isn’t even to push through a specific lie. Sometimes the aim is simply to sow distrust and poison debates. Whoever ends up believing no one anymore is easier to paralyze politically.

From troll factory to automatically generated audience

Classically, such an operation ran through paid staff in open-plan offices. These so-called troll factories managed hundreds of fake profiles with invented life stories. The accounts posted harmless content for months to build reach and credibility. Only afterward did the political messages arrive. The effort involved was high because every text had to be written by hand.

Language models have removed this bottleneck. A language model is a program that has learned from enormous amounts of text to predict fitting phrasing. This makes it possible to generate thousands of comments that all sound different and contain none of the typical telltale errors of a non-native speaker. Image generators supply profile photos of people who never existed. Voices, too, can be recreated, which enables faked video clips.

A common misconception is that such campaigns fail because of the technology. Mostly, they fail because of logistics. Accounts need phone numbers, access comes from conspicuous networks, and posts appear at similar times of day. Platforms therefore often detect campaigns not by their content but by their behavioral pattern. The technical term for this is coordinated inauthentic behavior.

Where the term shows up in reports and headlines

Major platforms and AI providers regularly publish reports on campaigns they’ve uncovered. These state which country an operation likely originated from, how many accounts were suspended, and which topics were targeted. OpenAI, for instance, has repeatedly described how groups from various states used its models to produce comments and articles. Such reports are the most common occasion on which the term appears in the news.

Regulation is also picking up on the topic. The European Union requires large platforms to assess risks to elections and document countermeasures. This matters for investors because it generates costs: review teams, detection systems, and possible fines. A company that overlooks a campaign on its platform risks headlines and proceedings.

In everyday life, people encounter the phenomenon without a label attached. A comment section that suddenly tips one-sided, or an account with no history that works on only a single topic, can be warning signs. It’s never certain, though, since real people sometimes behave similarly. An operation can usually only be proven once someone has access to the data behind the scenes.