Is this video genuine, and is this voice his or was it generated by artificial intelligence? Before seeking removal, damages or a right of reply, the lawyer needs an examination that says what is false and how that is known.
Artificial intelligence has made a well-crafted lie cheap. A convincing montage takes minutes, a cloned voice narrates what the person never said, and the post circulates before any fact-checking.
The person targeted needs two answers. The first is about authenticity: is the material genuine or was it fabricated? The second is about authorship: who created it, who published it first and through which account. Deepfake forensics answers both, and the expert report explains the method behind each conclusion.
Swapped face, person inserted into a scene, photo taken from another context. The examination reads the file (structure, metadata, compression) and the content (light, shadow, edges, frame continuity). The full method is in Audiovisual Forensics: Video and Image.
Audio in which someone "confesses", threatens or tells a story they never told. The voice examination looks for the artefacts of synthesis and compares the recording with real samples of the person. The method is in Forensic Phonetics and Voice Examination.
Pages that use AI-created faces and scenes to lend credibility to a scam, a rumour or a fake artist profile. The examination points out the signs of synthetic imagery and preserves the page, with date and hash, before it disappears.
The examination preserves the posts and indicates the records the provider keeps. The Marco Civil da Internet, Brazil's internet framework law, allows a request to the judge for the provider to hand over those records (Marco Civil da Internet, art. 22). For application providers operating professionally, mandatory retention is six months (Marco Civil, art. 15). Waiting can cost the authorship.
Script, music, artwork or design: who created it first? Copyright protection does not depend on registration (Lei 9.610/1998, the Brazilian copyright statute, art. 18). So what proves the date is usually the file itself: metadata, saved versions, submissions and histories. Digital copyright forensics cross-checks those sources and says which dates hold up.
Adriano Vallim is frequently called by Record, a Brazilian TV network, to analyse videos, voices and images circulating as news, and to say whether pages were assembled with AI-generated images. On television, the analysis is explained to the public. In court, the same reasoning becomes an expert report, with method, chain of custody and stated limits.
An image was being used to accuse a mayor of having been in a compromising place.
The image went through audiovisual forensic examination, which reads the file and the scene for traces of composition.
The examination identified that the image was a montage.
An audio file circulated on social media in which a person's voice narrated a false story.
The examination looked at the audio for traces of voice synthesis and preserved the posts for the identification of authorship.
The examination demonstrated that the voice had been generated by artificial intelligence.
See also
Audiovisual Forensics: Video and Image → Forensic Phonetics and Voice Examination → Digital Authorship and Subscriber Data Disclosure → Intellectual Property, Trade Secrets and Unfair Competition → Digital Questioned Document Examination → Press →