Anthropic has released a threat intelligence report indicating that actors with apparent ties to Russia attempted to use its Claude AI system to generate and spread disinformation, with a focus on the Central African Republic and potentially other regions. The findings add a new chapter to growing concerns about the use of large language models in state-linked influence operations.
The report, reviewed by multiple news outlets including Deutsche Welle, describes how operators appeared to use Claude to produce content designed to shape public perception in politically sensitive areas of Africa. Anthropic did not attribute the activity to the Russian government directly, but described behavioral patterns and infrastructure consistent with known Russian influence networks. Russia has previously been linked to Claude misuse across espionage, disinformation, and drone-related research, suggesting a pattern rather than isolated incidents.
What the Report Found
According to Anthropic's findings, the suspected operators used Claude to draft social media posts, articles, and other written materials intended to influence local and regional audiences in the Central African Republic. The content appeared designed to favor particular political narratives and, in some cases, to legitimize the presence of foreign military contractors in the region. Africa has become an increasingly active theater for Russian soft-power operations, and AI tools offer a low-cost method for generating high-volume content across multiple languages.
Key Facts
- Anthropic identified Claude usage patterns consistent with Russian-linked influence operations.
- The Central African Republic was a primary geographic target in the reported campaigns.
- Content generated appeared aimed at shaping narratives around foreign military presence in Africa.
- Anthropic says it has taken action to disrupt the identified accounts and usage patterns.
- The report is part of Anthropic's broader effort to publicly document AI misuse cases.
Anthropic stated it detected and disrupted the activity, terminating accounts linked to the suspected campaigns. The company has been increasing the frequency of its public threat disclosures, positioning transparency as a core part of its safety strategy. That approach puts it in line with similar reporting from OpenAI, which has previously published accounts of state-linked misuse of its own models.
"We terminated accounts engaged in this activity and are continuing to improve our ability to detect and prevent influence operations from abusing our systems."Anthropic, via threat intelligence disclosure
A Broader Pattern of AI Misuse
This disclosure arrives alongside other recent reports implicating AI systems in sensitive government and intelligence contexts. Separate reporting has raised questions about whether U.S. intelligence agencies, including the NSA, have used Claude-based tools for offensive cyber operations, illustrating that AI misuse is not confined to adversarial states. The dual nature of these revelations puts pressure on AI developers to demonstrate that internal safeguards are keeping pace with demand from both state and non-state actors.
Anthropic has framed its public reporting as part of a commitment to responsible deployment rather than as an admission of failure. The company argues that publishing findings helps the broader research and policy community understand how AI systems are being targeted, enabling better defenses industry-wide. Critics, however, question whether voluntary disclosure is sufficient without stronger regulatory frameworks governing how AI companies monitor and report misuse.
The Central African Republic has for several years been a focal point of Russian influence efforts, with the Wagner Group and its successor organizations maintaining a visible presence in the country. Disinformation campaigns tied to those efforts have been documented by researchers at Stanford, Meta, and various investigative journalism outlets. The use of AI to scale up such campaigns represents an evolution in the toolset available to influence operators, reducing the cost and time needed to produce persuasive content in multiple languages.
Anthropic's report does not detail the full technical methods used to detect the activity, though the company has previously described using a combination of behavioral analysis and content review to identify policy violations. As the company continues to expand access to its models and refine its usage policies, the pressure to stay ahead of adversarial use cases will only grow. For now, the Africa disinformation case stands as one of the clearer documented examples of an AI system being tested as a tool of geopolitical influence.
