Named by Edouard Bugnion as the best-in-class model for university-based research impact next to MDH. It combines technical network analysis, malware reverse engineering and investigative-journalism method, and uncovered the Pegasus spyware ecosystem used against journalists, activists, lawyers and politicians by identifying network infrastructure patterns, reverse engineering the software, and connecting the evidence to specific governments.

His question for EPFL: how would that combination of technical credibility and investigative output be built here?

Bugnion's second model, and the sharper one: a neutral university that studies internet phenomena empirically, project by project, without claiming to be the authority on the field. The UCSD group mapped the full supply chain of pharmacy spam operations by building measurement infrastructure and publishing peer-reviewed findings.

Applied here: instrument the Swiss information environment, election disinformation and platform hate speech, and act as an observatory rather than a detector.

Haitham Al-Hassanieh named this the most defensible moderation approach he knew of. When enough contributors independently annotate a post as misleading, with references, a note is appended for every viewer. It makes no claim that a post is false, it is human rather than automated, and it is transparent. Its weakness is scale: manual annotation is too slow when bots post thousands of times a minute.

His reading: fully automated detection at platform scale is not yet feasible, and human-in-the-loop systems are more trustworthy than machines issuing binary true or false verdicts.

The AI and Multimedia Authenticity Standards collaboration, a standing multistakeholder body convened by the three standards organisations to map what authenticity standards exist and where the gaps are. It is not EPFL, but Touradj Ebrahimi sits on it as Vice Chair while convening the JPEG group, which is how EPFL work reaches the international standards layer.

Its own outputs are papers, not this body: the first and second edition technical papers, the multimedia-authenticity policy paper, and the watermarking workshop report.

An association under Austrian law, around 600 member institutions, that governs the Time Machine project and builds historical knowledge infrastructure. Frederic Kaplan is its President. The argument for its place on an MDH map is his: sourced historical corpora are a counterweight to an information environment with no anchor.

Worth reading with the caveat the review found: timemachine.eu itself never mentions disinformation, misinformation, authenticity or verification. The MDH framing is Kaplan's, not the organisation's.

The cautionary case. Sabine Süsstrunk raises three failure modes in the provenance metadata standard: messaging platforms such as WhatsApp strip metadata on upload, so the label is gone before anyone sees it; editing tools apply the same AI-modified flag to removing a dust particle as to fabricating a photo outright; and a consensual self-made deepnude and a non-consensual harmful deepfake can carry identical labels. Touradj Ebrahimi reaches an overlapping critique from different examples.

Her institutional conclusion: EPFL should not sell detection or provenance guarantees to the public, because any such tool will be wrong at scale and carry the reputational liability.