TL;DR: We're building open cryptographic infrastructure to protect the authenticity of digital Holocaust testimony and archival material — before the window to do so closes. We work with the German Federal Archives, Holocaust memorial sites, and international research institutions. We're looking for technical and non-technical contributors (remote, volunteer basis) and BA/MA thesis candidates. Get involved here — a short message is enough.
Generative AI is making it increasingly easy to fabricate convincing images, audio, and video. The policy response so far has focused on two directions: deepfake detection – which is reactive, opaque, and at perpetual risk of being diminished by technical progress – and content credentials for newly created media, where proofs are embedded at the moment of recording.
Content credentials are genuinely important and will likely become standard. But they are structurally forward-looking: they certify new content from the moment of creation onward. That leaves decades of existing material – testimonies, photographs, historical recordings – without any comparable proof of authenticity.
This matters most where recorded testimony is irreplaceable – and where it is already a target for denial and distortion. Holocaust remembrance is the clearest case. As the last survivors die, recorded testimonies become the primary carriers of memory. These recordings were created long before content credentials existed. They will never carry embedded origin proofs.
The emerging asymmetry is this: as certified content becomes the norm, anything without a certificate risks being dismissed by default. A single expression of skepticism about its authenticity – "this could be AI-generated" – becomes sufficient to cast doubt on material that has been in institutional custody for decades.
There is a narrow window in which authentic recordings can still be reliably distinguished from synthetic fabrications. Once that window closes, retroactively establishing trust in existing material will no longer be feasible.
The Global Authentic Memory Initiative (GAMI) is a nonprofit building open infrastructure that allows archives and memorial institutions to create long-term, publicly verifiable proofs for their digital holdings. For any file in an archive, the system answers three questions:
The proofs are generated using cryptographic fingerprints, institutional digital signatures, and distributed timestamp anchoring. The files themselves never leave the institution. Verification is independently reproducible – no trust in any single platform, company, or state required.
We are a newly established nonprofit based in Munich, Germany, working in close collaboration with the Federal Archives, Holocaust memorial sites, and international research institutions. We have a working proof of concept and are preparing to enter the pilot phase with 3–5 institutions.
The team is currently small and growing fast – which is part of why we're writing this post.
Importance. Much of what defines democratic societies today – human rights, international law, the post-war institutional order – rests on a shared understanding of the past, grounded in testimony and historical evidence. If the authenticity of that evidence can no longer be independently demonstrated, what is at stake is not the credibility of individual archives, but the foundation on which these institutions were built. The infrastructure we are developing is domain-agnostic – applicable to war crimes documentation, journalism archives, human rights evidence, and scientific data – but our starting point is the domain where the stakes are highest and the institutional demand is clearest.
Neglectedness. Content credentials for new media are attracting significant investment from major technology companies and standards bodies – and rightly so. But as certified content becomes the norm, the pressure on existing material grows: anything without a credential is increasingly treated as suspicious by default. Retroactive authenticity infrastructure for historical archival holdings has, to our knowledge, no dedicated effort at a comparable scale anywhere in the world.
Tractability. The cryptographic primitives are well-understood and battle-tested. The core challenge is not research but engineering and institutional adoption – building usable tools, integrating them into heterogeneous archival systems, and getting institutions to actually use them. We already have written commitments from several major institutions. The bottleneck is capacity, not demand.
Time-sensitivity. The evidential strength of cryptographic timestamps depends on when they are created. Anchoring a file in 2026 – when indistinguishable synthetic fabrication is not yet widely achievable – is fundamentally different from anchoring it in 2030. This is one of the rare problems that becomes strictly harder to solve with each passing year.
We are looking for people who want to contribute – remotely, on a volunteer basis, at whatever level of commitment works for them. We're a nonprofit with limited funding so far, and we're being upfront about that.
Technical contributors. Cryptographic engineering, backend development, systems architecture. We're working on challenges ranging from scalable batch processing of large archival holdings to post-quantum signature migration, institutional key management, and building a public verification interface that non-technical users can use and understand.
Non-technical contributors. Research on funding opportunities, institutional outreach, communications, grant writing, and documentation.
Thesis projects. We offer supervised BA and MA thesis topics for students in computer science, information security, archival science, digital humanities, and related fields – with access to real institutional partners and data.
Connections and advice. If you know of relevant funding sources, institutional contacts, or people working on related problems, we'd be grateful for a pointer.
If any of this resonates, reach out at [email protected] or visit authenticmemory.org/join-us.
We're happy to answer questions in the comments.
