Pravda, RT poison Common Crawl LLM training data
Free public brief. The indicators, the detections with Splunk SPL already written (plus Elastic and KQL where they apply), and a Slack alert the next time this touches a vendor you run are in the app — free account, no card.
Russian state propaganda embedded in Common Crawl archive is being ingested by large language models, poisoning their outputs on geopolitical topics.
The DFRLab has documented systematic insertion of Russian state propaganda—including content from the Pravda network and RT—into Common Crawl, the public web archive that supplies training data for most open-weights large language models. An audit found approximately 40,000 English-language Pravda articles archived as of November 2025, up from 37 in November 2024. Testing with Llama 3.1 405B Base confirmed that models can reproduce RT's biolab disinformation nearly verbatim, demonstrating that the poisoning has already affected deployed systems. The mechanism is deliberate: Pravda configures its robots.txt and sitemaps to maximize crawling and indexing by Common Crawl's bot, ensuring rapid archival of propaganda content. Once baked into model weights during training, this content cannot be easily removed without expensive full retraining. The DFRLab also identified Chinese state-adjacent content from Glassbridge in Common Crawl, though JavaScript-based page architecture limited its ingestion into tested models. The threat is asymmetric: only 250 malicious documents can compromise even very large models, but defenders cannot surgically remove poisoned training data post-deployment. The risk intensifies as open-weights model development lags behind frontier systems and encounters Pravda material in the wild.
Mentioned in this report
Detection guidance
1 detection for this report is in the app — rules that match its indicators, converted to Splunk SPL, to Elastic where their fields map onto Elastic's integrations and to Microsoft Defender XDR KQL wherever Defender records the activity, plus YARA and Suricata. Three days of it free, no card.
Source reporting: https://dfrlab.org/2026/04/08/pravda-in-the-pipeline
What this brief leaves out
This page is the free tier: the write-up, the severity, the names. In the app the same report carries its extracted indicators, its detections (Splunk SPL already written, plus Elastic and Microsoft KQL where they apply), the actors and CVEs it names as live profiles, and the vendor research on the same campaign — the primary source the news paraphrased, cited and read directly.
It sits in a corpus of 13,854 reports from 146 sources, 616 of them written in the last seven days, and it grows through the day.
A new account starts with three days of all of it, no card, then keeps the government and CERT reporting free. Name the vendors you run and Slack hears about the reports that touch them — about forty a week that are yours, not the four hundred that are not.
Create a free account What it costs