Here’s something that should keep you up at night: AI is shaping how you vote, and Europe’s laws aren’t ready for it.

A new study from researchers at the Weizenbaum Institute (published in Communications of the ACM) makes a pretty alarming claim. Large Language Models — the same tech behind ChatGPT and Claude — are increasingly serving as information “gatekeepers” in ways that can subtly shift public opinion and voter behavior. And get this: current European legislation, including the EU AI Act and the Digital Services Act, barely addresses it.

The researchers found that these AI systems carry multiple biases. Their outputs are based on patterns in training data that preference specific worldviews and values. Worse, many AI systems are configured to reinforce a user’s existing biases or filter out certain types of content. It’s basically algorithmic confirmation bias at scale — and it’s happening inside the tools millions of people use to get information every day.

“The potential for Large Language Models to subtly influence political opinions and voter behavior poses a serious threat to public discourse and our democracy,” explains Stefan Schmid, Principal Investigator at the Weizenbaum Institute. He notes that this influence is often subliminal — making it nearly impossible for users to even recognize they’re being swayed.

Here’s what’s frustrating: the EU AI Act and DSA were designed to prevent “obvious” harm — things like illegal content, discrimination in high-stakes decisions, and explicit safety risks. But the subtle distortion of public discourse and democratic processes through bias in LLMs? That’s largely being ignored. The focus just isn’t there.

There’s also a consolidation problem. A handful of AI companies dominate the market, which means the diversity of perspectives in the digital sphere is narrowing. When everyone’s getting their information filtered through the same few models trained on the same few datasets, the risk of systemic bias becomes a risk to democracy itself.

The study suggests a broader regulatory approach is needed — combining content moderation, competition policy, value chain regulation, and technical design governance. Basically, we need to think about AI bias the way we think about media pluralism: it’s not just about one law, it’s about an ecosystem.

The question is whether Europe’s regulators are paying attention — and whether they’ll act before the next major election cycle. If you’ve ever wondered whether your vote is truly yours in the age of AI, this study suggests you have reason to wonder.