Where's a 'Red Scare' when you need one?
Russian propagandists turned one-third of relevant AI chatbot answers into Russian disinformation. Where's the panic?
Between 1917 and 1920, the United States underwent a "Red Scare" in response to the 1917 Bolshevik revolution that replaced the imperial monarchy with Soviet Communism. The hysteria was fueled in part by the fear that Russians were injecting pro-Russian talking points into the American media and other sources of information. The federal DOJ, state and local authorities, private vigilante groups, and anti-communist organizations often unfairly and with insufficient evidence targeted and persecuted labor activists, and sometimes deported immigrants.
A "Second Red Scare" gripped the United States between 1947 and 1957 in response to aggressive Soviet expansionism. Intense anti-communist paranoia swept the United States during these early Cold War years, characterized by widespread fear that Soviet agents had infiltrated American institutions, including the media and Hollywood. Also called "McCarthyism" after Senator Joseph McCarthy, the hysteria led to systematic efforts to identify and remove suspected communists from government, media, education, labor and the movie industry. Once again, the fear was that the Russians were injecting pro-Russian talking points into American sources of information.
Both "Red Scares" were characterized by a grain of truth magnified by fear and hysteria and resulted in unfair persecution, deportation, oppression, and, for lack of an older phrase, a “cancel culture” driven mainly by grandstanding demagogues.
In both "Red Scares," the degree to which pro-Russian talking points had infiltrated sources of information used by American citizens was exaggerated. The resulting oppression, censorship, and persecution did far more harm than the Russian propaganda.
Fast forward to 2025, and the level of pro-Russian propaganda in American sources of information is far beyond anything McCarthy could have imagined in his worst fever dreams.
Active readers already know about Russian influence operations in American information sources like the September 2024 Department of Justice indictment that revealed RT employees funneled $9.7 million to Tennessee-based Tenet Media to pay prominent right-wing influencers, the $6.8 million contract between Russian state media and conspiracy theorist Benjamin Swann uncovered through FARA filings, the Storm-1516 network that created sophisticated fake whistleblower videos targeting the 2024 election, and the seizure of 32 "Doppelgänger" domains that mimicked Western news outlets while disseminating Kremlin-directed propaganda aimed at reducing support for Ukraine.
Russian disinformation reaches American citizens through social media platforms, fake news outlets, viral videos, local influencers, paid internet trolls, botnets, and state-sponsored media outlets like RT and Sputnik.
The latest infiltration has been uncovered in AI chatbots.
According to the news site credibility rating organization, NewsGuard, a Moscow-based disinformation network called "Pravda" has infiltrated American AI chatbots with Kremlin propaganda.
By publishing some 3.6 million articles on 150 pro-Kremlin websites spreading 207 provably false claims, the "Pravda" network (which is also referred to by the excellent name "Portal Kombat," in some circles) was able to have enough of these articles picked up by AI companies’ automated data collection to strongly affect AI chatbot responses. Specifically, according to NewsGuard, "leading AI chatbots" — ChatGPT-4o, You.com, Grok, Pi, Le Chat, Microsoft Copilot, Meta AI, Claude, Google's Gemini, and Perplexity — repeated false narratives from the Pravda network approximately 33% of the time. Seven of the chatbots directly cited specific Pravda articles.
The articles exist explicitly to infect chatbots with pro-Russian propaganda. The websites barely have any organic traffic (fewer than 1,000 monthly visitors) but produce, on average, more than 10,000 articles per day. The tactic, called "LLM grooming," appears to flood the zone with quantity to target spiders and algorithms on the knowledge that chatbot companies are desperate to hoover up all data ever produced.
In other words, Russian disinformation networks — which exist to sway global public opinion in favor of specific candidates in elections, persuade citizens in democratic countries to embrace Russian interests (say, to oppose Ukraine in the Russia-Ukraine conflict), and sow division in Democratic societies — have discovered a "cheat code" for infiltrating the American mind: Flood the zone with massive quantities of pro-Russian articles, which the chatbots will gobble up and then spit out as fact when people ask Russia-related questions of the chatbots.
Given the degree to which Russia has successfully affected what Americans know and believe, shouldn't there be a third "Red Scare" — this time without the persecutions and hysteria, but instead a measured plan to weed this garbage out of our information streams?
Of course, the problem isn't about communism or even about Russia.
The problem is that anyone wishing to change the public opinion of any democratic nation can gamify our information streams. To the extent that Russia succeeds, others will follow—China, Iran, North Korea, the Republican Party, the Democratic Party, billionaires—you name it. Public opinion is for sale, and anyone can buy it.
All you have to do is use AI to churn out thousands of articles and post them online. To beat that effort, another malicious actor need only post millions. And the next propagandists billions. And in this contest for supremacy, the actual, authentic voices of legitimate and honest public discourse will be utterly buried under an avalanche of dishonest, manipulative garbage.
Sure, it can be tempting to blame the purveyors of LLM-based chatbots. And that's because we should absolutely blame the purveyors of LLM-based chatbots.
To be fair, most major AI chatbot companies say they’ve taken steps to deal with “LLM grooming.” They say they’re auditing and refining their training datasets to remove content from Pravda's network, enhancing their fact-checking mechanisms to better detect and debunk false narratives. And they’ve agreed to collaborate with watchdog organizations like NewsGuard to work on the problem.
But I feel like this is just a band-aid, or a fig leaf, or whatever the appropriate cliché is.
The problems isn’t Russian “Pravda” disinformation, specifically, but the gamability of public opinion generally through a loophole in AI chatbots and how they’re trained.
The AI chatbot companies want to steal everyone's content and make billions by offering up writers' and others' words, hijacking interest, and supplying the demand for that content while obscuring its source. But they haven’t demonstrated a broad plan for "LLM grooming" generally.
This is unacceptable. We don't want "Red Scare" hysteria and persecution. But we do want a nationwide focus on the problem of gamed outputs from AI chatbots, especially as LLM chatbots come to replace search engines as the main way people find stuff out.
And we need more transparency — much more transparency — about where our facts, information and opinions are coming from.
More From Elgan Media, Inc.
Drones are already transforming business
Why fast-learning robots are wearing glasses
Inside the sophisticated world of Mexico City’s haute cuisine
Where’s Mike? Oaxaca, Mexico!
(Why I’m always traveling.)
Absolutely agree - and if you get the underlying LLM you get the generative search engines built on top, and the agents built on top of that. Transparency should be something being demanded of AI developers, but they are slow to do that because of the implications for their copyright abuse. A dangerous period.
Absolutely agree - and if you get the underlying LLM you get the generative search engines built on top, and the agents built on top of that. Transparency should be something being demanded of AI developers, but they are slow to do that because of the implications for their copyright abuse. A dangerous period.