On 15 August, Mateusz was a guest on “Audycja Międzynarodowa”, the foreign-affairs programme of Polish Radio Trójka hosted by Agnieszka Lichnerowicz. The topic was LLM grooming: the deliberate manipulation of large language models with propaganda.

The mechanism is simple. Chatbots learn from the open web and increasingly search it in real time to answer questions. So whoever floods the web with content can shape what chatbots say. The best-known example is the Russian “Pravda” network. According to the American Sunlight Project, it published about 3.6 million articles in 2024 alone. These were written mainly for search engines and AI crawlers, not for human readers. A NewsGuard audit of ten leading chatbots found that they repeated false narratives from the network in about a third of the tested answers.

Mateusz explained why this matters for (cyber)security. More and more people use chatbots instead of search engines, often to get news or check whether something is true. If the models themselves are “poisoned”, disinformation reaches users through a source they trust. He also pointed to the responsibility of AI companies to filter their data sources and be transparent about them.

Link to the audition: https://www.polskieradio.pl/trojka/audycje/audycja-miedzynarodowa,11613/odcinek/aktualna-sytuacja-w-ukrainie-audycja-agnieszki-lichnerowicz,9e6620f9-5c48-4ada-9206-a5754195b7cd