Observed Signal · Jul 9, 2026 · Study/Investigation · Source: Meedia · Impact: 4/5 · Sentiment: Negative
Claude Under Pressure Over Russian Propaganda
A Berlin thinktank, Agora – Digitale Transformation, reports that Anthropic’s Claude cited material from a Russian disinformation network (the so‑called Pravda network) in answers to current-news queries. The case highlights a structural risk for large language models: when systems draw on open‑web sources, they can absorb and amplify manipulated content — a practice researchers call “LLM grooming.” The article notes prior findings from NewsGuard (2025) and a YouGov survey showing AI chatbots have become an increasingly important news source. Agora’s analysis also found differences between systems: ChatGPT and Google’s Gemini leaned more on journalistic sources, while Claude and Perplexity referenced more state or “private” sources. OpenAI disputed Agora’s implication that source selection was driven by commercial partnerships. The piece frames the issue as a growing regulatory and trust problem for AI vendors and the news ecosystem.
Study exposes systemic risk of LLMs amplifying disinformation, affecting news publishers, advertiser trust and prompting regulatory and liability concerns across the AI and media ecosystem.
Track Anthropic Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- Agora – Digitale Transformation found that Anthropic’s Claude referenced content from a Russian disinformation network (the “Pravda” network) in some answers.
- Agora researchers warn that mass-published, identical manipulated pages can be used to influence LLMs — a tactic termed “LLM grooming.”
- The analysis reported differences in source profiles: ChatGPT and Gemini cited mainly journalistic outlets, while Perplexity and Claude cited more state and private sources.
- OpenAI disputed Agora’s implication that citation patterns were driven by commercial partnerships.
- A Munich court recently confirmed that Google can be held liable for AI-generated answers in certain AI modes (as referenced in the article).
Connected Companies & Entities
5 Entities mapped“According to research by the Berlin thinktank Agora – Digitale Transformation, the Claude language model of the US company Anthropic referen...”
“Update: OpenAI rejects Agora’s assumption that the density of citations has anything to do with commercial cooperations....”
“A recent YouGov study shows that AI chatbots have become the second-most important news source after Google....”
“An earlier investigation by the organisation NewsGuard in 2025 concluded that several leading chatbots repeatedly picked up claims from this...”
“The study links ChatGPT’s frequent references to content from “Welt” and “Bild” to the partnership between OpenAI and Axel Springer....”
Ontology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
LLMs Aren't Brains — Expert on Anthropic's Claude
Anthropic said it has developed a new method that gives it deeper insight into how its AI models produce outputs. The article cites AI expert Will Douglas Heaven to interpret what those findings mean, and notes that Anthropic — maker of the Claude chatbot — is currently valued at nearly $1 trillion. The piece also references Anthropic's history of publishing provocative research (for example, on whether models can feel pain) and mentions that Claude can sometimes terminate conversations if it suspects users may be trying to 'abuse' the model. The original reporting is credited to James O'Donnell of MIT Technology Review and was published on t3n.de on 2026-07-15.
We Don't Communicate Just for Humans Anymore
A German PR columnist argues that generative AI systems (e.g., ChatGPT, Perplexity, Claude, Copilot) are increasingly acting as information gatekeepers, answering user queries directly and synthesizing content from many sources. This shift changes how visibility and reputation are formed: AI-generated summaries reward established, credible sources and deepen the strategic importance of specialist trade media and authoritative content. Operational PR tasks (monitoring, clipping, sentiment analysis, reporting) are being automated, but core PR work — setting topics, shaping narratives, building trust — remains essential. Communications must now optimize content not only for journalists and audiences but also for search, analysis, and answer systems to influence the composite picture these AIs produce. The column is authored by Christian Wick, a Swiss PR consultant and agency executive.
Journalism Faces AI Accusations and Detection Debate
The column reviews four near-simultaneous cases in German-language media where journalists or opinion pieces were accused of using AI: Tagesspiegel dismissed Editor-at-Large Stephan-Andreas Casdorff for having opinion texts produced by AI; Frankfurter Allgemeine Zeitung (FAZ) deleted a guest essay by Thuringia's minister-president Mario Voigt after it contained fabricated quotes; Axel Springer CEO Mathias Döpfner published a Welt column created with Google's Gemini and claimed not to have edited the suggested text; and investigator Stefan Weber alleged NZZ Deutschland's editor-in-chief Florian Eder used AI, based on the Pangram detector. The author argues AI-detection tools are statistical indicators, not proof, and warns that relying on AI as accuser undermines journalistic accountability and quality. The German Press Council has opposed a blanket AI-labeling requirement.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.