One of the most harrowing images from the Holocaust — a Nazi soldier pointing a pistol at a kneeling Jewish man before a mass grave — has long baffled historians. For decades, the location, victim, and perpetrator remained unknown or misattributed. But now, thanks to the blending of archival research and AI-driven facial recognition, historian Jürgen Matthäus claims to have identified the shooter.
The backstory, the process, the uncertainties, and what this means for historical justice all invite deeper reflection.
The Photograph: “The Last Jew in Vinnitsa” — Revisited
A Misattributed Icon
- The image, often referred to as “The Last Jew in Vinnitsa,” depicts a man kneeling, awaiting execution, with a mass grave behind him. For years it was thought to be from Vinnytsia, Ukraine.
- Matthäus’s research has established that it more likely was taken on 28 July 1941 in the citadel of Berdychiv (Berdychev) — roughly 150 km southwest of Kyiv.
- The shooter, according to Matthäus, is Jakobus Onnen, a former teacher, SS member, and murderer born near the Dutch-German border.
- While the certainty does not reach forensic crime-levels, the combination of AI matching, archival documents, and circumstantial links has made it a compelling identification claim.
Why It Matters
- This photograph is stark: it shows a direct, human-to-human act of killing, with no mediation by machinery or abstractions.
- By identifying the shooter, Matthäus restores agency: it ceases to be an anonymous act of terror and becomes a named crime.
- There is also an effort underway to name the victim — one among millions whose identities were erased during genocide.
- The project is part of a broader mission to restore names and stories to Holocaust victims, especially in the occupied Soviet territories where records were sparse and many remained unnamed.
How the Identification Happened: Methods, AI, and Human Judgment
Archival & Open-Source Investigations
- Matthäus spent years “digging in dusty archives,” examining troop movements, unit deployments, local wartime correspondence, and SS records.
- He collaborated with Bellingcat and volunteer researchers who helped source family photos, regional documents, and even private letters passed among descendants.
- A family member reported letters and images believed to belong to Onnen. While many original letters were destroyed, some family-held photos remained — crucial for comparison.
AI Facial Matching & Algorithmic Tools
- Using surviving photos of Onnen, AI-based facial recognition systems compared key facial landmarks against the shooter in the wartime photograph.
- The algorithmic match was strong (though not absolute). In forensic contexts, modern tools often claim 98–99 % matches; for historical, degraded photographs, matching is more challenging.
- Experts Matthäus consulted considered the match “unusually high” for a historical image.
- Still, the AI match alone was insufficient. It was buttressed by logistical, circumstantial, and archival evidence.
- Matthäus is careful to emphasize that AI is not a “silver bullet” — it is a new tool augmenting, not replacing, traditional historical methods.
Cross-Validation & Risks
- AI models are vulnerable to errors stemming from photo degradation, angle changes, resolution differences, facial aging, and stitching artifacts.
- The historical photograph has been reproduced and cropped in many variants; ensuring the correct frame and original image is essential.
- Biases in AI training data, misalignment, or misinterpretation posed risks — hence human oversight is still critical.
Broader Context: AI in Holocaust Research & Memory Work
AI Helping Restore Identities
This identification is not entirely unprecedented. Researchers and organizations have already used AI or machine learning to assist in identifying Holocaust victims:
- From Numbers to Names (N2N) is a project by Google engineer Daniel Patt which matches uploaded photos to Holocaust-era image archives. It returns top candidate matches for anonymous faces.
- Such tools have been used by descendants to trace relatives and help fill gaps in Holocaust victim registries.
- The United States Holocaust Memorial Museum, Yad Vashem, and other institutions have also experimented with AI-assisted image analysis in archives.
- These tools accelerate matching across thousands of photos, but they always depend on accessible reference images and high-quality scans.
Risks & Ethical Challenges
- Manipulation & misinformation: AI tools can be misused to create fake historical images or fabricate false claims. UNESCO has warned of the danger of rewriting history with deepfake content.
- Bias & data gaps: Many Holocaust-era records from Eastern Europe were lost or never recorded, making certain populations underrepresented in archival databases.
- Erosion of trust: Overemphasis on AI as infallible risks encouraging false certainty in ambiguous cases.
- Memorial distortion: Generative AI might produce “flashy” recreations that overshadow nuanced, grounded historical scholarship.
- Scholars urge that AI-based identifications must always be transparent, critically reviewed, and contextually grounded.
Implications: Justice, Memory & Historical Accountability
Naming the Perpetrators
- Identifying Onnen gives families, historians, and the public a clearer understanding of how genocide unfolded at the local level.
- It helps shift Holocaust memory from abstract statistics to personal histories of victims and perpetrators.
Empowering Descendant Communities
- Families of Holocaust victims, especially in Eastern Europe, often lack records, documentation, or closure. AI-assisted work can help reconnect survivors’ stories to names and faces.
- Such identifications can also support legal, reparative, or educational efforts.
Rethinking Historical Authority
- The use of AI introduces new tools into a discipline traditionally reliant on documents, human testimony, and archival research.
- Historians must balance excitement about capability with humility about uncertainty.
- AI can accelerate inference, but historians still must interpret evidence, weigh alternative hypotheses, and maintain methodological rigor.
Legal, Ethical, and Moral Questions
- What obligations exist to inform perpetrators’ descendants or communities?
- If a photograph is used or published, what consent rights (if any) do victims’ families have?
- Should there be public disclosure if AI makes a claim about identity — perhaps alongside margins of error or uncertainty?
FAQs — Common Questions About AI & Holocaust Photo Identification
Q1. How confident can we be in such AI-based identifications?
Confidence is always probabilistic. In this case, the AI match is strong but not definitive. Confidence increases when AI output aligns with archival records, circumstantial evidence, and human expert review.
Q2. Could the AI produce a false positive?
Yes. Degraded images, angle differences, facial aging, and poor quality reference images can mislead algorithms. That’s why corroboration is essential.
Q3. What happens to the victim’s identity?
Matthäus is also working to identify the kneeling man — by cross-referencing local Soviet-era community records, oral histories, and potential archival photos. But this is even harder, due to scarcity of surviving documentation.
Q4. Does AI replace traditional historical research?
No. AI is a tool, not a replacement. Historians must still judge sources, contextualize evidence, check for biases, and narrate interpretation.
Q5. Are there legal or ethical problems with identifying perpetrators decades later?
If the perpetrator is already deceased, legal prosecution is unlikely. But there are moral and familial implications — descendants may resist or embrace findings, and reputational issues can arise. Ethical handling, disclosure, and sensitivity are crucial.
Q6. What safeguards should be in place for AI use in historical work?
- Transparency about AI methodology, margins of error, and confidence levels
- Independent peer review before public claims
- Ethical review or oversight, especially when human communities are impacted
- Clear communication about uncertainty and ambiguity
Q7. Will AI make more such identifications possible?
Yes. As more digitized archives become available and reference datasets expand, AI can accelerate matching. But the quality and diversity of archival photos remain a limiting factor.
Q8. Does this work matter beyond Holocaust history?
Absolutely. It signals how AI could assist in restoring names in mass atrocity contexts (e.g. Rwanda, Armenia, conflicts in Africa), in truth commissions, or in uncovering historical crimes long buried.
In Summary
The claim that AI helped unmask the shooter in the “Last Jew in Vinnitsa / Berdychiv” photograph is a powerful example of how technology and history can intersect. But it also reminds us that:
- AI is a partner, not a replacement
- Evidence must always be triangulated
- Historical memory must be treated with humility, rigor, and respect
- Identifying individuals in atrocity photographs carries moral weight, not just scholarly triumph
By combining human perseverance, archival digging, and algorithmic insight, researchers are gradually restoring names and stories — turning shadows of history into tangible people, one face at a time.
Sources The Guardian
