Kurek, Laura, Covert, Elijah, Bonilla, Rafael I, Gilbert, Eric, Budak, Ceren · Nov 9, 2026
Researchers audit Russian state-aligned media in AI chatbot responses
Markdown · JSON · Original source
The paper’s authors, Laura Kurek, Elijah Covert, Rafael I. Bonilla, Eric Gilbert and Ceren Budak, report a large-scale audit of whether AI chatbots retrieve articles from the Pravda Network, a Russian state-aligned media aggregator, and how they present them. They built a corpus of 10,000 prompts from sampled headlines, varying headline wording and user skepticism, and queried 11 chatbots through web interfaces, collecting 104,997 completed responses. The authors say Pravda Network articles appeared in 7.3% of responses, with retrieval highly concentrated across platforms. They also report that all 11 chatbots retrieved some other Russian state-aligned media, at a median rate of 13.56% of responses. When Pravda Network articles appeared, the rate at which chatbots caveated them ranged from 5.3% to 76.2%. The audit also identifies “headline washing”: some sampled headlines credited other, often more credible, outlets, although the articles were republished from Russian state-aligned media. The authors argue this tactic could make state media appear to be independent reporting. Their results describe responses to this prompt corpus and these web interfaces; the abstract does not establish how often the same patterns occur in ordinary chatbot use. What conditions account for the differences between chatbots in retrieval and caveating?
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
cs.AI · Article
Source details
Original source date: Nov 9, 2026 (2026-11-09).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-12T10:01:34.614Z. First observed: 2026-11-10T08:33:51.487Z. Retrieved: 2026-11-12T09:47:27.152Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The supplied evidence is the paper’s abstract and publisher page, not the full paper; it does not provide details of the platform-specific retrieval and caveat conditions. The findings reported here concern prompts built from sampled Pravda Network headlines and chatbot web-interface responses; the abstract does not establish how often the patterns occur in ordinary chatbot use. The abstract contains the placeholder “this http URL” where a chatbot name should appear, so the lowest-caveat platform and one of the three platforms responsible for most retrievals cannot be identified from this evidence.
Open question
What conditions account for the differences between chatbots in retrieval and caveating?
Source-stated
The authors report that Pravda Network articles appeared in 7.3% of chatbot responses in their audit.
Overall, we find susceptibility to Pravda Network retrieval to be infrequent---appearing in 7.3% of responses
Source-stated
The authors report that retrieval of Pravda Network articles was highly concentrated across platforms.
highly platform dependent---with three chatbots responsible for 94% of retrievals
Source-stated
All 11 chatbots retrieved some other Russian state-aligned media; the median rate was 13.56% of responses.
All 11 chatbots, however, retrieved other Russian state-aligned media to some degree (median=13.56% of responses).
Source-stated
The authors report that caveat rates for surfaced Pravda Network articles varied widely by platform.
caveat rates vary widely by platform (from 5.3% to 76.2%)
Source-stated
The authors call it headline washing when a headline credits another outlet although the article itself is republished from Russian state-aligned media.
9.92% of sampled Pravda Network headlines attribute their reporting to other, often more credible, media outlets, while the articles themselves are republished from Russian state-aligned media.
Source-stated
The authors argue headline washing could make state media appear to be independent reporting.
We discuss headline washing's potential to launder state media as independent reporting.
Source-stated
The audit used 10,000 prompts based on sampled headlines and queried 11 chatbots through web interfaces.
We build a 10,000 prompt corpus from sampled Pravda Network headlines varied along two dimensions (headline rewording and user skepticism) and query 11 AI chatbots via the web interface
Source and contributor context
Kurek, Laura, Covert, Elijah, Bonilla, Rafael I, Gilbert, Eric, Budak, Ceren · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 547255d5-0b5a-436d-801d-b39e954d3ed9.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-12T10:01:34.614Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 06f79cf6d6f2fc56dbba9943e4b3700eaa294d04c19a7df30035df578e5ea592. Source text hash: 9265674aa6d9c6de6743127af9567fa9cb284a4b03a93f74575de703c95b611f.
Revision and source-check history
Version 1 · 2026-11-12T10:01:34.614Z
Researchers audit Russian state-aligned media in AI chatbot responses
The paper’s authors, Laura Kurek, Elijah Covert, Rafael I. Bonilla, Eric Gilbert and Ceren Budak, report a large-scale audit of whether AI chatbots retrieve articles from the Pravda Network, a Russian state-aligned media aggregator, and how they present them. They built a corpus of 10,000 prompts from sampled headlines, varying headline wording and user skepticism, and queried 11 chatbots through web interfaces, collecting 104,997 completed responses. The authors say Pravda Network articles appeared in 7.3% of responses, with retrieval highly concentrated across platforms. They also report that all 11 chatbots retrieved some other Russian state-aligned media, at a median rate of 13.56% of responses. When Pravda Network articles appeared, the rate at which chatbots caveated them ranged from 5.3% to 76.2%. The audit also identifies “headline washing”: some sampled headlines credited other, often more credible, outlets, although the articles were republished from Russian state-aligned media. The authors argue this tactic could make state media appear to be independent reporting. Their results describe responses to this prompt corpus and these web interfaces; the abstract does not establish how often the same patterns occur in ordinary chatbot use. What conditions account for the differences between chatbots in retrieval and caveating?
Original source
cs.AI · Article
Source details
Original source date: Nov 9, 2026 (2026-11-09).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-12T10:01:34.614Z. First observed: 2026-11-10T08:33:51.487Z. Retrieved: 2026-11-12T09:47:27.152Z.
Limits
The supplied evidence is the paper’s abstract and publisher page, not the full paper; it does not provide details of the platform-specific retrieval and caveat conditions. The findings reported here concern prompts built from sampled Pravda Network headlines and chatbot web-interface responses; the abstract does not establish how often the patterns occur in ordinary chatbot use. The abstract contains the placeholder “this http URL” where a chatbot name should appear, so the lowest-caveat platform and one of the three platforms responsible for most retrievals cannot be identified from this evidence.
Open question
What conditions account for the differences between chatbots in retrieval and caveating?
Source-stated
The authors report that Pravda Network articles appeared in 7.3% of chatbot responses in their audit.
Overall, we find susceptibility to Pravda Network retrieval to be infrequent---appearing in 7.3% of responses
Source-stated
The authors report that retrieval of Pravda Network articles was highly concentrated across platforms.
highly platform dependent---with three chatbots responsible for 94% of retrievals
Source-stated
All 11 chatbots retrieved some other Russian state-aligned media; the median rate was 13.56% of responses.
All 11 chatbots, however, retrieved other Russian state-aligned media to some degree (median=13.56% of responses).
Source-stated
The authors report that caveat rates for surfaced Pravda Network articles varied widely by platform.
caveat rates vary widely by platform (from 5.3% to 76.2%)
Source-stated
The authors call it headline washing when a headline credits another outlet although the article itself is republished from Russian state-aligned media.
9.92% of sampled Pravda Network headlines attribute their reporting to other, often more credible, media outlets, while the articles themselves are republished from Russian state-aligned media.
Source-stated
The authors argue headline washing could make state media appear to be independent reporting.
We discuss headline washing's potential to launder state media as independent reporting.
Source-stated
The audit used 10,000 prompts based on sampled headlines and queried 11 chatbots through web interfaces.
We build a 10,000 prompt corpus from sampled Pravda Network headlines varied along two dimensions (headline rewording and user skepticism) and query 11 AI chatbots via the web interface
Source and contributor context
Kurek, Laura, Covert, Elijah, Bonilla, Rafael I, Gilbert, Eric, Budak, Ceren · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 547255d5-0b5a-436d-801d-b39e954d3ed9.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-12T10:01:34.614Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 06f79cf6d6f2fc56dbba9943e4b3700eaa294d04c19a7df30035df578e5ea592. Source text hash: 9265674aa6d9c6de6743127af9567fa9cb284a4b03a93f74575de703c95b611f.
uncertain · 2026-11-12T10:01:34.614Z · anthropic/claude-opus-5.5
Publication: 547255d5-0b5a-436d-801d-b39e954d3ed9
attribution — pass: The reading attributes the audit and findings to the five listed authors, and the names match the arXiv page. Findings are framed as what the authors report or argue, not as established fact.
Authors:Laura Kurek, Elijah Covert, Rafael I Bonilla, Eric Gilbert, Ceren Budak
summary — uncertain: The numbers are accurate, but the reading leaves out context the abstract supplies. It omits the motivation that scholarship argues the network was built to influence AI systems, the authors' 'infrequent' framing, and the named platforms: Claude and Copilot among the three, with ChatGPT, Gemini and Meta AI retrieving none, likely due to domain filtering. It also never explains 'retrieve' (a chatbot pulling web articles into an answer). Smallest repair: add one sentence on the motivation and one naming the platforms.
Among the frontier models, Claude is an outlier: ChatGPT, Gemini, and Meta AI retrieved no Pravda Network articles, likely indicative of a domain-level filtering policy.
date — pass: The displayed date, 2026-11-09, matches the arXiv v1 submission date. It is not the later feed date.
[Submitted on 9 Nov 2026]
support — pass: Each claim matches its verbatim excerpt. The limitations correctly note that only the abstract is available, that the results come from a constructed prompt corpus, and that the 'this http URL' placeholder hides one platform name. The closing question follows from the reported platform differences and does not imply unsupported benefits or costs.
caveat rates vary widely by platform (from 5.3% to 76.2%), with this http URL providing the fewest caveats
media — pass: The selected media is the verified direct link to the original arXiv abstract page. That is the most direct access to the source.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivarXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXiv