Mateusz Łabuz and Matthias Schulze co-authored OpEd titled “KI-Hintertür in die Demokratie” (German version available online). The OpEd was published by Tagesspiegel Cybersecurity Background.

They pointed out that large language models reproduce Russian disinformation themes in 33 percent of the cases observed by researchers. For that reason, Russian disinformation actors are increasingly trying to influence the results generated by LLMs. This strategy has already been called “LLM Grooming”. As they argue, although the disinformation biases showed by LLMs might be an outcome of deliberate actions of threat actors, the effect of this LLM grooming can also be amplified by the normalization of disinformation and lies in politics.

The OpEd can be read here.

Photo credits: Tagesspiegel Cybersecurity Background