Today, memorial institutions dealing with the authoritarian past and mass atrocities committed during it face a mnemonic paradox. On the one hand, there has never been more high-quality information available about genocides, such as the Holocaust, due to the digitization of pre-digital materials and new digital content produced by the institutions. On the other hand, ensuring that this information reaches the audiences — and is not lost in the growing volume of misleading content that floods online spaces — is becoming harder, and it contributes to the troubling findings about the low levels of awareness and knowledge about the crimes of National Socialism as captured by the recent Claims Conference’s cross-country survey.

A core reason for this paradox is the growing role of algorithm- and artificial intelligence (AI)-driven systems as gatekeepers of genocide memory. Given the unprecedented amount of information about the past available online, it is simply impossible for humans to curate it. An immediate consequence is that this task is delegated to automated systems, such as search engines, which organize information sources and retrieve them in response to user queries. In the case of queries regarding genocides, however, search algorithms — as well as social media algorithms — do not necessarily prioritize content from heritage institutions. It limits the visibility of the resources accumulated by institutions and forces them to compete for users’ attention with content providers, which often have better search optimization strategies and more resources for self-promotion.

Right now, however, the collective memory ecosystem faces another change. The emergence of generative AI and its applications, such as ChatGPT, changes how information about the past is discovered and created. In a 2025 survey we conducted with Victoria Vziatysheva, a rising human-AI interaction scholar, we found that 18.6% of US respondents and 12.7% of German respondents reported using generative AI chatbots to acquire information about the Holocaust (based on a sample of 1,000 respondents per country). Unlike search engines, however, chatbots do not guide individuals to external information sources; they directly provide answers. As a result, even if the answer includes links to other information sources, a chatbot user does not need to engage with them, as the requested information is already provided.

The challenge of generative AI in the context of remembering the National Socialist past is not limited to the risk that it will make memorial institutions’ presence in online environments obsolete in the long run. Research shows that answers from generative AI applications about genocides are often far from accurate and prone to propagating stereotypes. Concerningly, this biased representation of the past can be triggered by rather conventional user requests, leading to a phenomenon which we describe with my colleague, Elizaveta Kuznetsova from Weizenbaum Institute, as a routine Holocaust distortion – i.e., a non-prompted misrepresentation of the past arising from the inherent aspects of generative AI, such as its tendency to hallucinate.

Another risk of generative AI is that it can be easily used to produce historical materials that may appear authentic to a non-expert but are, in essence, fake. While there are promising uses of AI-generated historical materials by activist and artistic groups, for instance, the “Pedanterie – the Auschwitz Laundry” by the Konrad Wolf Film University of Babelsberg, it is contrasted by the growing volume of non-authentic content dealing with the Holocaust in non-expert-curated digital platform spaces. It is often unclear what motivates platform users to create such content (especially since many of them remain anonymous), but it likely spans a broad spectrum of intentions, from self-expression to monetization to genocide denial. In the end, the growing volume of history-related AI slop poses a risk not only of preventing individuals from finding quality content but also of undermining trust in historical facts.

It is worth noting, however, that these technological developments enable not only risks but also substantive possibilities. So far, I have focused on the former, but it would not be unfair to omit the latter. Just as guiding individuals away from memorial institutions, algorithmic systems can guide people towards them; the outcome depends on the underlying design and functionality principles of the system and the eagerness of the developers to acknowledge the importance of their work for the collective memory (as well as the presence of regulation focused on preventing the risks of technology for this under-regulated area). Similarly, generative AI can facilitate new forms of media transmission, such as more responsive digital avatars of genocide survivors or personalized historical lessons from chatbots equipped with high-quality information, provided these technologies are developed and adapted responsibly.

The important question, however, is how to limit the risks and realize the possibilities. It is not an easy task, but it is not impossible. Memorial institutions like KZ-Gedenkstätte Neuengamme have been raising their voices in recent months to draw attention to the risks of AI-generated Holocaust distortion in online environments, and research groups led by scholars of AI and Holocaust, such as Victoria Grace Richardson-Walden, Todd Presner, or Tobias Ebbrecht-Hartmann, are increasingly looking into the subject. It is important, however, that attention to the problem is translated into action, be it improved safeguards from AI developers, new forms of regulation to account for the emerging forms of genocide distortion, or new education techniques for developing historical literacies at a time when both facts and evidence can be easily fabricated by AI.

Memorial institutions have invaluable expertise that can provide guidance in developing all these solutions, provided they are willing (and able) to engage with other memory stakeholders. In a world where it is so easy to produce fake historical materials, the importance of institutional actors who ensure the integrity of historical knowledge cannot be overstated. Finding new ways to ensure that the importance of authoritative historical knowledge is recognized by the general public, stakeholders, and AI systems can help resolve the mnemonic paradox I noted at the beginning of this contribution.