Anthropic exposes AI operation that DRC journalists never knew existed

Disinformation surrounding DRC-Rwanda relations has weakened the diplomatic and public standing of both countries

Originally published on Global Voices

This post is part of Global Voices’ October 2026 Spotlight series, “Big Tech and Authoritarianism.” In this Spotlight, we explore how Big Tech facilitates and hinders authoritarian governments and practices, and how these choices influence the Global South. You can support this coverage by donating here.

In September 2026, Anthropic, an artificial intelligence (AI) development and research company, exposed a network of fake news sites targeting the Democratic Republic of the Congo (DRC). The investigation revealed that large-scale, AI-powered disinformation campaigns have emerged at a time of conflict when trust in the news is already at an all-time low and resources are severely limited.

This revelation rang alarm bells for African media watchdogs. The network coordinated around 70 AI-powered fake news sites and generated at least 318 articles on the eastern DRC conflict, relations with Rwanda, and contested mineral resources. The operation was highly robust.

On X, coordinated accounts amplified content from the AI-run sites, mimicking legitimate media outlets and rewriting articles for different audiences.

According to Anthropic’s report, the LKM Company, a French advertising agency serving clients with strong interests in the DRC-Rwanda conflict, was behind it all. They found no evidence of direct government involvement.

When Global Voices approached Hobab Watukalusu, a fact-checker, editor, and head of the training and research department at Soma Média Lab in Kinshasa, the capital of the Democratic Republic of the Congo, he shared:

Nous n’avions pas identifié cette opération dans toute son ampleur avant sa révélation. Cela montre les limites auxquelles nous sommes confrontés face à des campagnes capables de produire, à très grande échelle et dans des délais très courts, des contenus qui peuvent paraître crédibles et s’insérer rapidement dans l’écosystème informationnel.

We hadn’t recognized the full extent of the operation before this revelation. This highlights our limitations when dealing with campaigns that can rapidly generate large-scale, credible-looking content and quickly gain a foothold in the infoscape.

He added that local journalists on the front lines are struggling to keep up as new technologies make it easier than ever to facilitate massive disinformation campaigns.

Fact-checking obstacles

DR Congolese fact-checkers face a litany of obstacles in slowing disinformation flows across regions. Hobab outlines the fundamental constraints:

L'accès encore difficile à l'électricité, à une connexion Internet stable et abordable, ainsi que le coût élevé de la vie constituent des contraintes réelles pour les professionnels de l'information.

Ongoing difficulties in accessing electricity, stable and affordable internet connections, and high living costs pose real-world constraints for journalists.

In the DRC, where journalists are already overwhelmed by covering an active military conflict, analyzing potential industrial-scale, AI-generated content is virtually impossible.

Joël Kaseso, another journalist covering disinformation narratives, told Global Voices:

Des ressources existent, mais la plupart sont payantes. Ce qui limite notre capacité à tout analyser dans ce contexte tendu où les relations diplomatiques et sécuritaires avec les pays voisins, notamment le Rwanda, ne sont plus au bon fixe.

Although resources exist, most are not free. Amid this tense environment, where diplomatic and security relations with neighboring countries, particularly Rwanda, are strained, this financial barrier limits our ability to conduct a full analysis.

Kaseso has also identified a simple but critical methodology that most journalists use to navigate this situation: analyzing the author’s intentions, the source’s profile (anonymous accounts versus identifiable influencers), and amplification patterns. However, this requires time and tools that many newsrooms don’t have.

The challenge extends far beyond detection. Hobab explains:

Détecter qu'un contenu a été généré ou manipulé à l'aide de l'intelligence artificielle ne constitue que la première étape. Il faut ensuite déterminer si ce contenu est trompeur, identifier sa source, comprendre comment il se diffuse et, surtout, établir dans quel objectif il a été produit. C'est cette chaîne complète de vérification qui demande du temps, des compétences et des ressources.

Detecting AI-generated or AI-manipulated content is only the first step. The next steps involve determining whether the content is misleading, identifying its source, understanding how it spreads, and, above all, establishing why it was created in the first place. This comprehensive verification process requires time, skill, and resources.

The resource gap is pivotal. It ultimately determines who controls the narrative during a crisis. While an AI-powered operation can generate 318 targeted articles in just a few days, a genuine newsroom must verify its sources, contact those involved, corroborate images, and provide context, all of which can take weeks.

When disinformation becomes a weapon

The Anthropic report revealed another major threat jeopardizing investigative reporting in the DRC. Accusations of disinformation can themselves become weapons of censorship. Hobab points out a troubling precedent:

Il existe malheureusement des cas où les accusations de désinformation ou de diffusion de fausses informations ont été utilisées dans des procédures visant des journalistes. Le cas de Stanis Bujakera est particulièrement révélateur de cette tension.

Unfortunately, accusations of spreading disinformation and fake news have been weaponized in legal proceedings against some journalists. Stanis Bujakera’s case is a prime example.

In September 2023, DR Congolese journalist Stanis Bujakera was arrested for spreading false information. In March 2024, the court subsequently sentenced him, raising serious concerns among press freedom organizations.

The current state of affairs complicates fact-checkers’ work and could stifle journalists’ efforts. When asked how media personnel can combat genuine disinformation without giving governments a new tool to silence their critics, Hobab stated:

Pour éviter que les politiques de lutte contre la désinformation ne deviennent elles-mêmes des instruments de répression, il faut mettre en place des mécanismes de régulation réellement indépendants et transparents. Le risque est de transformer une réponse à un problème informationnel en un nouveau problème pour la liberté de la presse.

To prevent anti-disinformation policies from becoming instruments of repression themselves, establishing truly independent, transparent regulatory mechanisms is vital. Solutions to information-related problems risk becoming another threat to press freedom.

Diplomatic impact

Although Hobab remained cautious about claiming direct causality, he noted that the 318 AI-generated articles had a tangible impact on the region. As Joël Kaseso put it:

Oui, la désinformation autour des relations RDC-Rwanda a beaucoup fragilisé la position diplomatique et publique congolaise. Cela a alimenté une haine envers les Rwandais et renforcé des faux récits sur le Rwanda.

Yes, disinformation surrounding DRC-Rwanda relations has significantly weakened the DRC’s diplomatic and public standing, fueling hatred towards Rwandans and reinforcing false narratives about Rwanda.

Heavy exposure to biased, manipulated, and highly emotional narratives, often framed as “true patriotism,” makes it harder to distinguish verified information from propaganda. Hobab also points out specific political consequences: measures affecting universities in AFC/M23-controlled areas and restrictions that Burundi announced on entry from M23-controlled areas.

Ces décisions illustrent à quel point le contexte sécuritaire et informationnel régional peut avoir des répercussions concrètes sur la vie des populations.

These decisions illustrate just how much the regional security and information landscape impacts people’s lives.

In Rwanda, the situation seems much the same. Manzi (a pseudonym), a Rwandan journalist, explains:

Au Rwanda, nous constatons que la désinformation pro-RDC circule largement. Les citoyens rwandais sont exposés à des récits qui dépeignent le Rwanda comme occupant militaire, ce qui renforce les divisions. Mais un défi demeure. Ici, distinguer entre désinformation malveillante et critique légitime devient politiquement chargé.

In Rwanda, DRC disinformation circulates widely. Rwandan citizens are exposed to narratives portraying Rwanda as a military occupier, deepening divisions. However, one challenge remains. Here, distinguishing between malicious disinformation and legitimate criticism becomes politically charged.

Encouraging initiatives

Despite the challenges, DR Congolese journalists are actively working to build effective resistance techniques. Hobab mentioned Kinshasa News Lab, a newly formed space where fact-checkers, media professionals, and media literacy educators meet to discuss the causes of disinformation and possible solutions. Joël Kaseso also noted partnerships producing measurable results:

Les collaborations avec l'IFCN et Meta nous ont beaucoup aidés à combattre les faux récits autour des relations RDC-Rwanda. Le signalement des comptes propagandistes et des trolls a vraiment fait diminuer la désinformation.

Our collaborations with the IFCN [the International Fact-Checking Network] and Meta have helped combat false narratives surrounding DRC-Rwanda relations. Reporting propaganda accounts and trolls has really helped reduce disinformation.

He also lists freely available tools that do not require any expensive licenses: InVID We Verify, Google Image Search, Yandex, Hive Moderation, and Zero GPT to detect AI-generated content.

Training journalists in fact-checking has enabled them to critically analyze online content and distinguish fact from fiction. However, according to Hobab, this training remains insufficient. He warns:

Le défi n'est plus seulement d'apprendre aux journalistes à vérifier l'information classique. Il faut désormais les préparer à comprendre les mécanismes de l'IA générative, les réseaux de diffusion, les campagnes coordonnées et les nouvelles formes de manipulation.

The challenge is no longer just a matter of teaching journalists how to fact-check conventional information. We must now help them understand how generative AI works, how information spreads, how coordinated campaigns operate, and new forms of manipulation.

Instrumentalizing fact-checking poses yet another threat. Fact-checking is the ultimate safeguard for verified information. However, if it instead becomes a tool in the political, economic, or institutional sphere, it loses much of its credibility.