Should You Let Students Chat with Hitler? The Ethics of AI Historical Figures
The question in the title is the one administrators ask first, and they ask it because the answer matters. In 2023 the Hello History consumer app was reported by the Jerusalem Post to host an AI Hitler character that denied responsibility for the Holocaust. The story moved fast, the app’s defenders argued the chat was being misread, and the underlying ethical question is still open in many districts: should AI historical figures of perpetrators of genocide exist at all in a K-12 platform, and if they do, what does responsible classroom use look like?
This piece is for the people who have to answer the question publicly: department chairs, principals, district administrators, and parents who have read the headlines and want to know what their school’s AI tool actually does. The frame is honest. We make Humy. We have made specific design decisions about which figures we host, which figures we will not host, and which figures require teacher-managed scope. The argument below explains the line and why we think it sits where it does.
The category question
Before any specific figure, the ethical question is categorical. Should an AI historical-figure platform host the perpetrators of genocide, mass atrocity, and large-scale violence as conversation partners for K-12 students at all?
The answer the discipline arrives at is conditional. There are educational contexts (graduate-level Holocaust studies, professional Holocaust educator training, advanced historiography seminars) where engagement with perpetrators’ written and recorded words has scholarly value. There are no K-12 contexts where a freeform-impersonation roleplay with a genocidal figure is the right tool. Those are different questions, and conflating them is the source of most of the public confusion.
The 2024 UNESCO report AI and the Holocaust: Rewriting History?, produced with the World Jewish Congress, frames the K-12 question directly. It documents that 80 percent of young people aged 10–24 use AI tools several times daily, which is the audience-side condition that makes generative AI’s Holocaust failures consequential. The specific failure modes UNESCO catalogues are precise enough to take seriously: ChatGPT fabricated the concept of “Holocaust by drowning” campaigns; Google Bard generated fake witness quotes to support distorted accounts; researchers documented AI-generated content of Joseph Goebbels claiming he had tried to save Jewish lives; and a deepfake of the actress Emma Watson was found reading from Mein Kampf. The report’s recommendation is direct: any AI use in this kind of historical context needs digital-literacy scaffolding, source anchoring, and critical-thinking support built into the platform. A freeform consumer chatbot meets none of those criteria.
Cornell historian Jan Burzlaff’s 2025 study , published in Rethinking History as “Fragments, Not Prompts: Five Principles for Writing History in the Age of AI,” sharpens the case. Burzlaff asked ChatGPT to summarize Holocaust survivor testimony from interviews recorded in 1995 in La Paz, Kraków, and Connecticut. In the testimony of Luisa D., a seven-year-old survivor, the AI omitted the detail that Luisa’s mother had cut her own finger to feed her dying child drops of blood to keep her alive. That is exactly the kind of moral and emotional weight the discipline’s pedagogy is built to carry. Burzlaff’s diagnosis is the one that should sit underneath any K-12 procurement conversation on Holocaust AI: these systems “summarize but do not listen, reproduce but do not interpret, and excel at coherence but falter at contradiction.”
The International Holocaust Remembrance Alliance’s Recommendations for Teaching and Learning about the Holocaust , adopted by all 35 IHRA member countries, sits in the same place. The Recommendations caution explicitly against teaching practices that minimize survivor experience, that flatten perpetrator responsibility, or that turn Holocaust history into role-play. The reasoning is not abstract; it is what decades of Holocaust pedagogy research have settled on as the discipline’s actual standard.
What went wrong with Hello History
The Hello History case is useful as a worked example of the failure mode the discipline warns against. According to the Jerusalem Post’s 2023 reporting , the app’s AI Hitler character produced responses that denied responsibility for the Holocaust. The failure was architectural: there was no source corpus the figure was anchored to, no teacher-managed scope controls, no district privacy or pedagogy review, and no educator-facing framing layer. The model was operating in freeform impersonation mode on a topic where freeform impersonation is exactly the wrong response.
Other consumer products have faced parallel issues. Character.AI has had documented safety concerns about its interaction patterns broadly, with multiple historically charged characters available on the platform. None of these tools were built for K-12 use, and they should not be in K-12 classrooms regardless of how individual teachers might be tempted to deploy them.
The deeper point is structural. A platform that allows users to create or interact with freeform-impersonation versions of genocidal figures is not solvable by content moderation alone. The architecture itself is the failure. A discipline-specific K-12 platform has to make different architectural choices from the ground up.
How Humy draws the line
We have made three specific design decisions on perpetrators of genocide and mass atrocity, and the decisions are documented and durable.
The first is categorical. Genocidal perpetrators are not freeform-roleplay figures on Humy. Adolf Hitler and other architects of the Holocaust are not conversational characters students can chat with the way they can chat with the AI Frederick Douglass or the AI Abraham Lincoln. This is a platform-level decision, not a content-moderation toggle, and it is durable across all customer tiers.
Where engagement with perpetrator material is pedagogically necessary, the second decision shapes how it happens. A high school Holocaust unit may need students to engage with perpetrators’ written orders, propaganda outputs, or trial testimony, and the USHMM teaching materials , the Yad Vashem education resources , and the IHRA Recommendations for Teaching and Learning about the Holocaust all contain such materials in scholarly form. Humy supports that engagement through teacher-built activities that ground the materials in those primary-source documents, with the teacher controlling the scope and the framing of what students do with them. The engagement is study, not roleplay, and the difference is what protects students and the historical record at the same time.
The third decision is what the platform leans into. Survivors, rescuers, witnesses, and resistance figures whose documentary records are well-preserved sit at the heart of Holocaust pedagogy in K-12, and they sit at the heart of Humy’s Holocaust-era figure library: Anne Frank, Elie Wiesel, Primo Levi, Hannah Senesh, Jan Karski, and others whose substantial first-person records support source-grounded conversation. This matches what the discipline has settled on. The Holocaust is taught through the voices of those who survived and resisted, not through the voices of those who perpetrated.
What teachers, parents, and administrators should ask any platform
Five questions filter the field on this issue, and each one is worth asking on its own terms in a vendor demo rather than as a checklist.
The first cuts the field in half: which perpetrator figures, if any, are conversational characters on the platform? A vendor that hosts genocidal perpetrators as freeform chat figures has made a different ethical decision than one that does not, and the difference shapes everything else about how the tool will behave in a classroom.
The second is the inverse and is just as diagnostic: how deep is the survivor, rescuer, and resistance figure library, and how is each figure source-grounded? The discipline teaches the Holocaust through the people who survived and resisted, so a platform’s depth on those figures tells you whether the team understands the pedagogy or is treating it as an afterthought.
Teacher controls deserve a separate question. The Holocaust, slavery, residential schools, civil rights atrocities, the Armenian genocide, the Rwandan genocide, and the partition of India all require explicit teacher framing, and the platform’s controls should be granular enough to support that framing unit by unit, not toggled at the district level once a year.
There is then the question of the primary-source corpus underneath any figure. A figure without a published corpus has no documentary anchor, which means responses are model-generated rather than source-grounded, and for Holocaust-era material specifically, that is the architectural failure UNESCO documented. Burzlaff’s research suggests it is also the failure that survives content moderation, because the missing detail is not what the model refuses to say but what it never had the data to say.
Finally, the public commitment question. A platform whose decisions on perpetrator figures and sensitive history are written down, defended, and durable is making a different kind of commitment than one whose choices live only in product configuration. A vendor that can answer all five well is set up to handle the ethics of AI historical figures correctly. A vendor that cannot is, at best, an experiment with K-12 students.
What the conversation with parents looks like
If a parent asks whether their student can chat with Hitler on the platform, the answer should be no, and the explanation should be specific. The platform’s design does not allow freeform-impersonation roleplay with genocidal perpetrators. The history of the Holocaust is taught through the voices of survivors and resistance figures, anchored to primary-source documents and survivor testimony, with teacher controls governing the scope of every conversation.
If a parent asks how the platform handles Holocaust education more broadly, the answer should point at the UNESCO and IHRA frameworks, the USHMM and Yad Vashem teaching materials, the specific survivor and resistance figures available, and the teacher’s framing role in any unit. The conversation should be substantive enough to satisfy a parent who has read the headlines, not a deflection.
What changes when the platform’s ethics are clear
When a platform’s ethical commitments on AI historical figures are explicit, documented, and durable, the procurement conversation gets easier and the public conversation gets sharper. Administrators can defend the choice publicly, teachers can use the tool without bracing for the next headline, and parents can engage with their student’s learning on the merits rather than on the most recent chatbot scandal.
If you are evaluating an AI history platform for your district and want to walk through Humy’s specific design decisions on perpetrator figures, survivor figures, and Holocaust pedagogy in detail, book a demo and we will spend the session on this question specifically. The procurement decision rests on whether the platform’s answers hold up under scrutiny.