Setting the boundaries for AI in Holocaust education

A student raises her hand and asks a question. The woman on the screen pauses, blinks and answers. The classroom falls quiet. For a moment, it feels as if a Holocaust survivor is sitting with them, listening and remembering.

I have sat in those rooms. I have watched that silence settle. It is powerful, and it is also fragile.

Because the survivor is not there.

The students are speaking with an AI-powered hologram of Sonia Warshawski, a Holocaust survivor whose testimony has been carefully recorded and preserved. Developed through The Blue Card's innovative hologram program, she is the first survivor to participate in this initiative, which is now expanding to include more of their voices as we teach the next generation.

Warshawski, who is 100 years old, is far away, living her life in the Midwest. Students interact not with a system that recreates her presence, but one built to preserve her words. The system is listening, not remembering.

As artificial intelligence and generative AI rapidly enter classrooms across the nation, the number of Holocaust survivors continues to decline, with about 196,000 remaining worldwide, most now in their late 80s and 90s. Holocaust education remains inconsistent, with many students reporting little to no exposure and nearly half of Americans unable to name a single concentration camp, according to the Claims Conference.

The question is no longer whether we will use AI in Holocaust education, but how.

At The Blue Card, which serves Holocaust survivors in need, I work with survivors every day. I know how carefully they choose their words and what it costs them to tell their stories. They are not offering material to be expanded or interpreted. They are bearing witness to what they lived.

That is why the system we built is deliberately nongenerative. It does not invent, interpret or fill in gaps. It retrieves only what a survivor actually said.

Ask Sonia: What did you see when you arrived at Auschwitz?

Ask Sonia: What happened to your family?

Ask Sonia: What do you remember about the death marches?

If the answer exists, she responds with her own words. If it does not, the system pauses. That silence is honest. These limitations are not flaws, but safeguards.

Generative AI works differently. It predicts language by determining what should come next. When trained on survivor testimony, it does not retrieve memories, but constructs responses. It can produce statements no survivor ever made, delivered in a voice that sounds real.

That is not acceptable in Holocaust education, where accuracy is an obligation. At a moment when Holocaust knowledge is already eroding, we cannot afford tools that blur the line between what was said and what could have been said.

This is not theoretical. Recently a 15-year-old Syosset High School student and his father were arrested after the teen allegedly drew a swastika in a bathroom, and investigators found chemicals the father bought for his son to make rockets, officials said. For Holocaust survivors, incidents like this are retraumatizing, triggering anxiety and painful memories of the very hatred that nearly destroyed them. The need to educate is urgent, and local.

Artificial intelligence will shape the future of education, including Holocaust education. It can expand access and bring students closer to voices they might never otherwise hear.

But AI should not speak in place of survivors. Systems that pause, that stop, that refuse to answer, are doing something essential. They protect the boundary between what is known and what is not. That boundary is where truth lives.

This guest essay reflects the views of Masha Pearl, the executive director of The Blue Card.