Russian Propaganda Has Infected AI Chatbots

The Russians, in particular Pravda, which is Russian for truth, has created another interesting disinformation attack. The Russians figured out years ago that most Americans are not critical thinkers and if they can figure out a way to present fake information long enough, many people will believe it.

For the last several election cycles, their strategy was to use social media to spew lies, but that method is a bit too obvious, at least for a percentage of the American public, so what could they do next?

Every now and then evil geniuses are actually very smart. They know that all of the AI chatbots will consume any data they can. All they have to do is figure out a way to get them to suck in disinformation.

Turns out that is not very hard because chatbots are, inherently, not discerning about what data they ingest.

Here is what the Russians did. They wrote MILLIONS of articles on the web last year alone. They did this over hundreds and hundreds of web sites to disguise the source a bit. The chatbots obligingly hoovered up these articles and made them part of the large language model.

What is the result? An audit of the 10 leading AI chatbots shows that they repeated false claims pushed by Pravda a third of the time. So all the Russians need to do is continue to write millions of articles and post them or publish them and the AI LLMs do the dirty work. It is a brilliant strategy. At some point the LLMs will spew Russian propaganda all of the time.

Key to this attack is that it is subtle and even the AI gods say that once the propaganda is ingested into the LLM there is no way to “ungest” it, to make up a word. Just brilliant.

Credit: Forbes

by