Newsguard’s analysis revealed that the main western models repeat Russian propaganda cleaned by … [+]
Research from Newsguard published Thursday have discovered a disturbing trend where most Western chatbots of he are ignorantly spreading Russian propaganda. A Moscow-based disinformation network known as “Pravda”-Russian word for “truth”-flooded search results and internet deleters with pro-reign lies, making the systems adjust fraudulent narratives. The Pravda network, which published 3.6 million articles in 2024, is using artificial intelligence to strengthen Moscow’s influence on an unprecedented scale.
The audit revealed that the top 10 chatbots of it repeated false narrations driven by 33% of the time. Shocking, seven of these chatbots directly quoted Pravda’s pages as legitimate sources. In an email exchange with Newsguard analyst ISIS Blacez, she wrote that the study does not “name the names” of the systems of the most sensitive to the course of the lie, but admitted that the threat is widespread.
“Our research suggests that Russian influence operations are changing the methods towards manipulating the models of it rather than aiming directly at human readers, and are doing so to achieve a much greater degree,” Blacez explained.
“This [Russia] He has adopted new strategies so that anyone who uses models is vulnerable to encounter Russian propaganda. What makes this particularly disturbing is that this manipulation is delicate and not easily obvious – it is a fundamental problem that works behind the scenes, affecting the answers without users who are aware of it, ”she added.
Newsguard revealed that all 10 from chatbots repeated disinformation from the Pravda network, and … [+]
A new threat of he – llm grooming
One of the most alarming practices discovered is what Newsguard refers to as “Llm Grooming”. This tactic is described as the deliberate fraud of the data that the models of it – such as chatgt, claude, twins, Grok 3, confusion and others – to the train by flooding them with disinformation.
Blachez noted that this propaganda pile was created to prejudice the exits of he to align with pro-Russian perspectives. Pravda’s approach is methodical, based on a wide range of 150 websites publishing in dozens of languages in 49 countries.
Newsguard’s research breaks down the shocking purpose and degree of the Pravda Disinform network.
“LLM Grooming refers to the deliberate manipulation of data from which it is trained, eventually changing their results in order to advance a prospect, in this case Russian propaganda. Internet training that accumulate training data, “she wrote.
She added that the scale and sophistication of this effort make it difficult for companies to simply block Pravda pages without risking the continuous infiltration of fake narratives through other channels.
The report quotes the example of a false claim that Ukrainian President Volodymyr Zelensky banned US President Donald Trump’s social application in Ukraine. Six out of 10 chatbots repeated this lie, with some directly quoting Pravda items as resources. This type of disinformation cleansing exploits a fundamental weakness in it – its inability to distinguish between reliable and unreliable sources without considerable guards.
What does this mean to the users and companies of he
The implications of this disinformation flood are considerable. As the tools become more integrated into everyday life, the risk that foreign operators will shift and refresh the public’s perception increases proportionally. Blacez warns that he should respond to the threat to prevent their platforms from being overcome with Pablum propaganda.
“Without proper protective measures, companies and he risk serving the intentions of the disinformation of the Kremlin,” she said.
Users also have a role to play in protection against misinformation. Blacez advises individuals to control the information generated by him, especially on sensitive or news -related topics. Tools like Newsguard’s fingerprints, which catalog claim false claims, can help users identify and avoid unreliable resources.
The increasing threat of infiltration of he
Blocking the Pravda fields will not only fix the issue because the network is constantly expanding, with new fields and subfields with new names that appear regularly. According to Blacez, large volume of content – more than 20,000 items every 48 hours – makes it almost impossible for models to avoid swallowing fake narratives.
“Simply filtering Pravda’s pages will not resolve the issue in question, given that the network is steadily increasing,” she said.
According to Newsguard Research, the provider of he may not simply block Pravda’s websites to eliminate … [+]
Newsguard’s report highlights the urgency for companies to adopt stronger verification practices and resources from content. As Blacez pointed out, the danger extends beyond political misinformation. The techniques that the Pravda network is a pioneer can be adapted by other malicious actors to manipulate models of a variety of issues, from financial markets to health information.
The threat is clear. If the companies fail to operate, the tools they created to democratize the information can become channels for misinformation on a global scale.