Bringing more archives into our verification network, researching AI forgery of historical records, and making the risk known while we still have time
Holocaust records are more accessible than ever, because so many more are available online than in the past. But when you come across a photo on social media of a scene from Auschwitz with a convincing story to go with it, in the age of generative AI, how do you know that it’s real? This contributes to an even more sinister problem: being able to dismiss real records as fake, because of how widespread ‘AI slop’ is online. Denying evidence of a documented atrocity becomes easy.
Authentic Memory is a small German non-profit (gemeinnützige UG) founded by passionate students from LMU. Archives sign their files with a hardware key, and we anchor those signatures using independent timestamping software. Any person can check that a file is exactly what the archive signed, and that it existed in that form on that date. So far, we've anchored 68,555 files across 4 institutions.
Our current grants run until April 2027, and cover the anchoring infrastructure, the core technology and partner institution onboarding. They fund only a minimal level of outreach and no research beyond the core programme. This fundraiser adds to those grants within the same period: more outreach, a separate research project on how AI forgery threatens archives, and work to make verification more visible to archivists, journalists and the public.
Anchoring more institutions sooner. Every archive that joins before April 2027 is onboarded under our current grants. We have committed to our funders to onboard 10 institutions by then (though our own goal is at least 15), plus a pipeline of 50 more. Since signing archives one at a time is slow, we are also approaching associations and umbrella organisations, with the aim of being able to create one agreement that can cover several collections.
Visibility. Our current grant only funds one pilot workshop. With this funding we will run further training for archivists, journalists and educators, give talks at gatherings of archive leaders, and write recommendations for funders of digitisation projects for how to make anchoring an integral part of the projects they support.
Research. As a separate project, we will keep a dated, public record of what generative tools can fake for each type of media, so claims such as "this was anchored before convincing fakes existed" rest on tangible evidence. We will publish our threat model, including the limits of our system, and test our public verifier with non-specialists such as students and journalists, to make sure our verification results are easily understandable by the general public.
Our existing grants cover the anchoring infrastructure, the technology and a minimal level of outreach. We will report how the money raised here is spent in the financial section of our annual report. We plan to spend it on the following, in order of priority:
$10K (minimum): outreach to new institutions beyond what our grants allow, meaning travel, onboarding meetings and conference presence. This is the most time-sensitive part, because institutions that join before April 2027 are onboarded under our current funding.
Up to $25K: adds the research project and publishing its results.
Up to $50K: adds the training sessions, talks and the funder recommendation.
Beyond $50K: a reserve for costs our current grants don't cover, such as unexpected expenses during onboarding.
Founder & Director:
Pablo Toussaint
Tech:
Peter Van Garderen: Chief Technology Officer
Sumit Vekariya: Software Engineer
Jay Sojitra: Software Engineer
Chijioke Ugwuanyi: Software Engineer
Institutional Partnership:
Emma Yanai
Communications:
Rosa Bauer: Head of Communications & Impact
Hannah Kennedy: Communications Intern
Operations and Research: Sonakshi Chauhan
Academic advisory board:
Ali Sunyaev (Vice President & Professor of Information Infrastructures, TU Munich)
Jacob N. Shapiro (John Foster Dulles Professor of International Affairs, Princeton University)
Anja Ballis (Professor of German Didactics, Holocaust Education, LMU Munich)
Luciana Duranti (Professor Emerita of Archival Science & Diplomatics, University of British Columbia)
Memorial sites and archives are cautious about cryptographically signing their records, but we have gained early trust from our pilot partners. Records from the Dachau Concentration Camp Memorial Site and the NS Documentation Centre are already anchored in our network. National archives and genocide memorials in Europe and Asia are now onboarding.
We work with institutions and research organisations including Starling Lab for Data Integrity (Stanford/USC), EHRI-ERIC, memorial archives and InterPARES. Our academic advisory board connects us to TU Munich, Princeton University, LMU Munich and the University of British Columbia. Together, this integrates our infrastructure with archival practices and builds on the research base that already exists.
Our work is funded by Google.org, Germany’s Federal Commissioner for Culture and the Media (BKM) and the EVZ Foundation (Remembrance, Responsibility and Future). Each one carried out its own due diligence before funding us.
The most likely failure is time. Archives move slowly, and some institutions won't be ready to sign before our current funding cycle ends. Therefore, much of the anchoring will happen after that date, and without new funding we could not continue to onboard these institutions free of charge. The extra outreach this project pays for is meant to bring as many institutions as possible to the point of signing while onboarding is still covered. Whatever happens, the files already anchored stay verifiable.
Authentic Memory is funded by Google.org and the German Federal Government Commissioner for Culture and Media (BKM), with grants totalling €400,000. It is also funded by the Foundation EVZ with €50,000. The funds cover the anchoring infrastructure, the standard, pilot projects, and the onboarding of partner institutions.
There are no bids on this project.
Comments
Sonakshi chauhan
about 18 hours agoUpdate: In October, Google.org and Germany’s Federal Commissioner for Culture and the Media announced their support for our work, and the news was picked up by 51 publications, including dpa. Since then, more archives have reached out to us. Timing matters because generative AI gets better at faking documents, photos, and video every few months. A record anchored today can be proven to have existed before convincing fakes of it were possible, but once fakes are everywhere, that proof is much harder to establish. Questions and feedback are very welcome.