Who benefits from the crime ?

Anthropic has lifted the veil on a new dimension of information warfare: the industrial use of artificial intelligence to create fake media outlets, manufacture digital identities and give the appearance of spontaneous public debate to carefully orchestrated political campaigns.

In its report entitled Detecting and countering misuse of AI: September 2026, Anthropic says it uncovered and disrupted several operations that misused its Claude model for manipulation, surveillance, cybercrime and political influence.

One of these operations, labelled GTG-54002, is of particular interest to Central Africa and relations between the DRC and Rwanda. Anthropic attributes it to LKM Company, a France-based advertising and digital marketing agency.

A factory of fake news outlets

According to Anthropic, the operation relied on around 70 fake news websites, designed to look like legitimate and independent media organisations.

These sites were accompanied by roughly 70 X accounts directly corresponding to the fake media outlets, as well as a network of more than 250 fake commenting and amplification accounts.

Altogether, the operation reportedly published at least 8,913 articles in around 20 languages, targeting audiences across six continents.

Claude was used in two main ways: to generate entirely new articles and to rewrite stories from genuine media outlets in ways that reflected a desired political narrative.

Artificial intelligence therefore enabled the operators to rapidly transform existing information into multiple versions tailored to a particular country, target audience and political message.

The DRC among the main targets

Anthropic identifies four areas that were especially targeted by the network: the United States, Brazil, France and the Democratic Republic of Congo.

Its detailed analysis of the Congolese component is particularly revealing.

Across the network, Anthropic identified 318 articles specifically focused on the DRC. According to its researchers, these publications generally supported the Congolese government’s position and focused especially on two issues: the continuing tensions between the DRC and Rwanda, and regional agreements concerning minerals.

The operation did not stop at publishing articles. It also sought to create the appearance of genuine public opinion around them.

During the first weeks of the campaign, Anthropic observed that the overwhelming majority of fake profiles following the network’s accounts were associated with the DRC. A page devoted specifically to Congolese news even became, at one point, the most widely shared and most popular account in the entire artificial network.

Researchers also noticed that an X account presenting itself as belonging to a Congolese civilian “digital army” followed several accounts linked to the operation.

Articles on the DRC-Rwanda conflict published almost simultaneously

One of the clearest signs of coordination appeared on September 11, 2025.

On that day, several fake websites belonging to the network published almost identical articles on the conflict between the DRC and Rwanda within only a few minutes of one another — in a window of roughly three minutes.

The underlying content was similar, but the tone was adjusted depending on the regional audience each fake outlet claimed to serve.

At the same time, links to these articles were circulated in a coordinated manner by the associated X accounts.

This pattern helped Anthropic’s investigators identify what they describe as “coordinated inauthentic behavior” — coordinated activity designed to make an organised operation appear to be the spontaneous expression of multiple independent media outlets and social media users.

An operation prepared from France

Investigators were also able to reconstruct part of the technical infrastructure.

The domains used by the fake media outlets had been registered from France over a period of roughly ten weeks in mid-2025. The various sites shared common technical infrastructure, allowing Anthropic to connect dozens of supposedly independent media brands to a single operator.

Most of the social media accounts linked to the operation had been created in June and July 2025.

Among the fake media brands named by Anthropic were Naija Pulse, Axum Voices, Jambo Journal, Zion Pulse, Echo Berlin, The British Daily, Russian Way and Commonwealth Post.

This diversity of names and geographical references was precisely intended to create the illusion of numerous independent information sources spread across the world.

A company able to switch sides depending on the client

The report also provides an important clue for understanding the nature of the operation.

According to Anthropic, the network did not have a fixed political ideology. It could promote completely different political positions depending on the country and, above all, the interests of its clients.

Anthropic therefore describes LKM Company as part of what it calls “influence-as-a-service.”

In other words, this was not necessarily an organisation campaigning for one political cause of its own, but rather a commercial operator capable of selling clients a complete influence infrastructure: fake media, AI-generated articles, false online identities and artificial amplification across social networks.

Who paid for the content favourable to Kinshasa?

This is where an important distinction must be made.

Anthropic says it found indications that the campaign may have reflected the interests of one or more clients concerned with the DRC-Rwanda conflict.

However, the company makes clear that it was not able to independently identify the clients who commissioned the content.

Most importantly, Anthropic explicitly states that it found no evidence proving that any government directed the operation.

It would therefore be premature, on the basis of this report alone, to state that “the Congolese government used Claude against Rwanda.”

What the report does establish is nonetheless significant: a France-based commercial agency operated a covert fake-media network, a substantial part of whose Central African output generally supported the DRC’s positions in its disputes with Rwanda.

The key question therefore remains unanswered: who were the clients financing this part of the operation?

An impressive volume, but apparently limited impact

The actual impact of the campaign also needs to be put into perspective.

Although nearly 9,000 articles were produced worldwide, Anthropic says that most of them appear to have generated little genuine engagement from real users.

The company therefore places the operation at Level 2 on the Brookings Institution’s Breakout Scale: the content circulated within the network of fake media outlets and controlled social accounts, but Anthropic found no evidence that it had successfully penetrated the wider public debate on a significant scale.

Anthropic also says it intervened early enough to prevent the network from building a large authentic audience. The accounts and organisation involved were banned from its services, and new detection mechanisms were introduced.

A new dimension of information warfare

The case goes far beyond relations between Kigali and Kinshasa.

In the highly polarised context of relations between the DRC and Rwanda, this technology therefore introduces a new actor onto the battlefield: the industrial manufacture of a public opinion that does not, in reality, exist.

Anthropic’s report offers a rare glimpse into some of these mechanisms from the inside.

But it also leaves open the most politically important question: who, in the DRC-Rwanda context, paid LKM Company to produce and distribute these hundreds of articles?

Until that question is answered with documented evidence, caution remains necessary regarding the exact identity of those behind the campaign.