Research dispatches, detection benchmarks, and forensic analysis from AFIP's content authenticity labs.
Our initial analysis of detection accuracy against the latest language models, including methodology notes and false positive rates across 8,400 test samples.
Examining how AI voice clones replicate — and fail to replicate — micro-variations present in natural human speech across extended utterances.
New findings on identifying which generative model produced a given synthetic image, with implications for provenance tracking.
Testing the persistence of content authentication watermarks across multiple re-encoding and compression cycles.
Exploring methodological parallels between pathological examination and forensic analysis of digital content.
Preliminary framework for establishing forensic evidence standards when AI-generated content is introduced in legal contexts.