Germany Funds AI to Evaluate Political Statements

Critics fear that a tool designed to combat disinformation could itself introduce political or institutional bias if its sources, training data and assessment criteria are not sufficiently transparent. Disputes during the Covid pandemic demonstrated how quickly contested claims can become entangled with broader political arguments over what constitutes reliable information.

The issue therefore extends beyond the technical detection of deepfakes. It touches on freedom of expression, media freedom and the ability of political actors to challenge prevailing narratives without being inaccurately categorized as spreading disinformation.

The consortium is coordinated by the German Research Center for Artificial Intelligence (DFKI). Its partners include the Technical University of Berlin, Otto von Guericke University Magdeburg, Ernst Abbe University of Applied Sciences Jena, Deutsche Welle, the German Press Agency (dpa) and Gretchen AI GmbH.

The dpa alone is receiving €262,000 ($301,000) in project funding and is responsible, among other things, for helping to define editorial requirements.

Deutsche Welle, Germany’s publicly funded international broadcaster, is also participating. It is expected to test the technology and explore its potential journalistic applications.

German Chancellor Friedrich Merz has meanwhile taken a striking position on freedom of expression online. Discussing a possible requirement for social-media users to post under their real names, Merz rejected objections based on free speech and argued that a liberal society must also protect its citizens against false information.

A state-funded AI assistant designed to identify and contextualize potentially misleading political content inevitably brings those two principles into tension: protection against manipulation on the one hand and the freedom to challenge accepted narratives on the other.

Who Decides What Counts as Disinformation?

The Fake-O-Meter project starts from the premise that conventional fact-checking is valuable but often arrives only after misleading content has already spread. Its researchers are therefore also examining preventive approaches intended to make citizens more resistant to anticipated disinformation.

That approach, often known as “prebunking”, raises a different set of questions from conventional fact-checking. Instead of responding only to a specific factual claim, it can involve identifying recurring manipulation techniques or narratives before they spread widely.

For such a system, the choice of sources and the way conflicting evidence is weighted become particularly important.

Alternative for Germany (AfD) lawmaker Nicole Hess has criticized the project on these grounds. A state-funded Fake-O-Meter without published criteria or effective oversight does not protect democracy, she argues. “Anyone who wants to measure truth must first disclose the standards being used”, she told German media outlet Nius.

Technology connected to the research project is already being used ahead of the election to Berlin’s House of Representatives on 20 September. The service offered by the state of Berlin is called Die Gretchenfrage, a German expression referring to the decisive question.

Users can submit suspicious social-media posts to the service. The system examines images and videos for signs of synthetic generation or manipulation before comparing claims with official information, established sources and professional fact-checks.

The Berlin authorities stress that the AI does not deliver a definitive judgment about whether a political post is true or false. Instead, it is supposed to present evidence, sources and, where necessary, uncertainty so that users can form their own opinion.

The service has nevertheless attracted considerable interest. Berlin’s State Returning Officer reported more than 17,000 queries between its launch on 26 August and 7 September.

The technology behind Die Gretchenfrage was developed by Gretchen AI, a spin-off from DFKI. DFKI describes the Berlin deployment as connected to the research continuing under the Fake-O-Meter project, which began in May 2026 and is scheduled to run until February 2029.

The proximity to an election makes transparency particularly important. Users need to know not only what conclusion an AI-assisted system reaches but why it reaches it, which sources it relies upon and how uncertainty or conflicting evidence is handled.

There are also questions about data protection when users submit social-media material that may contain information about people who themselves have not interacted with the service.

Good Intentions, Serious Risks

The developers of Die Gretchenfrage and the broader Fake-O-Meter project stress that their technology is meant to support independent judgment, not replace it. That makes the criteria behind its assessments all the more important.

Published standards, traceable sources, independent scrutiny and a clear means of challenging errors would make it easier to establish whether such systems are genuinely neutral in practice.

There is an important distinction between detecting whether a photograph has been digitally manipulated and interpreting a political “narrative”. The former can often be treated as a technical forensic problem. The latter may involve context, competing interpretations and political judgment.

The government wants to protect citizens from misinformation and manipulation. But when a state-funded system moves from detecting fabricated images to assessing political narratives, another question arises: how can citizens know that it is not itself shaping the debate through opaque assumptions, source selection or classifications?

In such a sensitive area, good intentions are not enough. Transparency is essential.