When a photo or video may be synthetic, detection is not only a technical question. It is mostly a governance issue: who can verify the result, who can audit the method, and who has the capacity to challenge mistakes?
Today, deepfake detection is often treated as a black-box service: media goes in, a score comes out. That may be convenient, but it is not enough for contexts where decisions affect journalism, elections, public administration, legal proceedings, or citizens’ rights. Closed, proprietary systems make it difficult to assess bias, robustness, data handling, failure modes, or whether a detector still works once content has been compressed, resized, reposted, or stripped of metadata.
This talk argues that open-source detectors are not just useful for researchers and scientists in the field. They are a prerequisite for accountable governance. Public institutions need tools they can inspect. Regulators need benchmarks they can reproduce. Journalists and civil society need methods that can be questioned rather than simply accepted. And Europe, if it wants digital sovereignty in synthetic media verification, needs more than procurement access to proprietary, opaque scores.
In this talk, we will share lessons from our open-source work on detector benchmarking, social-platform degradation, synthetic-media datasets, and provenance testing. The aim is to show why open detectors matter: not because they solve deepfake detection once and for all, but because they make the limits, trade-offs, and assumptions of verification visible.
