AI Disinformation & Zersetung
- Richard Murff
- 7 days ago
- 3 min read
The new surgical age of fake news is here
It looks like social media is set to be the 8-Track of news media - a popular innovation that wasn’t as convenient as advertised and ill-suited for the medium. It wasn’t that long ago when legacy outlets were complaining about the young migrating to unchecked social media for their news. Now the upstarts are feeling the pain: usage peaked just after the pandemic and has been sliding since. Per a 2026 Reuter’s Interactive Digital news Report, some 10% of people globally – and 17% of the 18-24 cohort – now get their news via AI chatbots.
On the surface, the sheer certainty of chatbots are comforting. AI removes the random, lone-wolf human craziness that makes coherent narratives impossible to control. The alternative may be worse. An AI chatbot make a narrative very easy to control. The danger is just who is controlling it.
As intelligence agencies around the world are discovering, there are several ways of doing this. The first is simple the direct influence of state run media. Control the chatbot in question, and you can control its answers. Ask China’s DeepSeek how to make and deliver bomb, it won’t tell you. Nor will it admit that anything untoward happened at Tiananmen Square in 1989 or any other year. It will note that you asked, however.
What the Chinese bots do, generally is put guardrails around the subjects that it doesn’t like or want to talk about. The human problem is that everyone knows that the system is rigged. I once asked someone about state media in Nicaragua, and the reply was as straight forward as it was hopeless: “Oh, we know it isn’t true. But it is all people see, and risky to speak out.”
Less obvious, and more effective, is indirect control of the chatbots that a party has no control over. It’s called Data Poisoning. Current AI models assess the accuracy of facts largely on the number of times a fact or narrative appears online. So assessment of what it scrubs isn’t based on wide-context understanding or anything we’d call critical thinking. In practice its AI version of Joseph Goebbels’s If you repeat a lie enough times it becomes the truth logic.
It is on this premise that a Kremlin active measures – what we call disinformation – unit recently used AI to create and deploy some 200,000 SEO optimized websites to manipulate AI sources during Germany’s massive rearmament push. Once you’ve set the perimeters, a stand alone websites can be created in seconds and human help. The universe of sites are optimized not for human traffic, but for other bots. In the olden days of social media, you actually needed followers and forwards to spread the crazy. Today it doesn’t really matter how much human traffic the dummy sites gets: If the chatbot sees it 200,000 times, it must be true.
If you want seeds to take root, break-up the solid ground before planting.
It’s true that a Russian active measures unit hacked the DNC ahead of the 2016 Trump v. Clinton election, but most forget the operation hacked the Republican National Committee as well. The operation soon pivoted tactics to be semi-obvious about one breach in order to sew chaos. And chaos, not electing a particular candidate to the White House, was always the goal. Why? Because if you want seeds to take root, break-up the solid ground before planting them. And in the case of Russia, the seed they wanted to grow was called Zersetzung or, less exotically, internal rot.
Western intelligence agencies often share - or leak – reports to friendly reporters in sympathetic news outlets to get a narrative out. Social media complicated the picture with a galaxy of random, dopamine triggering rage-bait. Unlike social media, AI takes the random human crazy out of the equation with a very calculated one to seed zersetzung. One that it doesn’t merely manipulate the information, but it alters the lens through which people see process events around them.
The Field Guide:
So how do you get around that? Unfortunately, you don’t, really. That having been realized, its hard to express how much good it will do your sanity and perception of reality to simply know that this sort of manipulations is taking place. Remember, if it isn’t on the internet, then that AI chatbot or model doesn’t know it. That is the model on which it has built the universe. And while it’s getting harder to convince the kids, the internet is a terrible model for reality.
I’ve written here before that the confidence curve is a trap, and it is. While its counter-intuitive, but simply admitting to yourself that you don’t have all, or most, of the relevant information will help you make better decisions. As well as help you correct course when you do cock it up.
