The transition from the textual dominance of the "Gutenberg Galaxy" to an ocularcentric era characterized by synthetic media fundamentally alters the ontology of historical evidence. As generative artificial intelligence becomes capable of fabricating hyper-realistic audiovisual content, the digital public sphere faces a profound epistemic crisis. The boundary between authentic historical records and algorithmic hallucinations is dissolving. While Digital Humanities has traditionally excelled in text-centric hermeneutics, the field currently lacks adequate methodological frameworks to address the multimodal nature of contemporary disinformation. This paper posits that addressing this "blind spot" is not merely a technical challenge but a foundational imperative for the humanities. We argue that the university must reclaim its role as a guarantor of truth by developing sovereign "Infrastructures of Truth."

The current landscape of media verification relies dangerously on "epistemic outsourcing." The advanced tools required to detect synthetic content are almost exclusively proprietary and operated by the same technology corporations that facilitate its creation. This reliance on corporate "black boxes" presents a dual threat to academic integrity: it compromises data sovereignty by processing sensitive research data through opaque commercial APIs and creates dependency on unexplainable algorithms, rendering verification scientifically non-reproducible. If the academy relies on commercial entities to distinguish signal from noise, it surrenders its critical agency.

In response, this contribution proposes the implementation of Local Inference Clusters as a decentralized and university-governed infrastructure for forensic media analysis. Drawing on a case study of detecting "Biological Incongruence" in political discourse, we demonstrate how open-source Large Language Models and multimodal neural networks can be deployed within the secure perimeter of research institutions. This approach ensures strict adherence to ethical standards and GDPR compliance while allowing researchers to inspect the internal reasoning of the models.

Furthermore, this project operationalizes the concept of "hybridity" by dissolving the rigid barrier between computational forensics and humanistic inquiry. We posit that an infrastructure of truth cannot function solely as a technical filter; it must operate as a participatory knowledge space. By transparently visualizing the dissonance between audio signals and facial micro-expressions, we provide a scaffold for public digital literacy. This approach moves beyond binary "true/false" labeling, offering instead a "pedagogy of verification" that empowers citizens to recognize the material traces of synthetic manipulation. Thus, the research infrastructure serves not only to detect falsehoods but to cultivate a more resilient, critically engaged public sphere capable of navigating the synthetic age.

We frame this architecture as a renewal of classical Quellenkritik (source criticism) for the algorithmic age. This methodology, which we term Forensic Hermeneutics, empowers humanists to move beyond the passive interpretation of cultural artifacts to the active verification of their material authenticity. By shifting from external corporate dependencies to resilient internal computing capacities, Digital Humanities can provide a critical public value. This infrastructure of engagement offers independent, transparent, and scientifically rigorous tools to civil society, ensuring that the power to discern reality remains a public good rather than a privatized commodity.