Sin categoríaDeepfakes and Audiovisual Evidence: Challenging AI-Generated Video and Audio in Spain

For decades, a video or a voice recording produced in proceedings enjoyed almost automatic credibility. Image and sound seemed to speak for themselves. That premise has broken: today any user with a home computer can generate, in minutes, an audiovisual sequence in which a real person says or does something they never said or did. These are the so-called deepfakes, and their irruption forces a rethink of how audiovisual evidence is handled in the criminal process.

What exactly a deepfake is and why it is an evidentiary problem

A deepfake is synthetic content generated with artificial intelligence models that replicate faces, voices and gestures with high fidelity. The evidentiary problem is twofold. First, well-made fabrications no longer show the gross inconsistencies that allowed traditional montages to be detected. Second, and more serious, their mere existence generates what Anglo-Saxon doctrine has called the liar’s dividend: whoever appears in an authentic and compromising recording can claim it is a fabrication, sowing a doubt the prosecution will have to dispel.

Practical key: the file produced in the proceedings is almost never the original. It is usually a copy forwarded through WhatsApp or downloaded from a social network, recompressed and stripped of its metadata. That circumstance, by itself, already weakens its evidentiary value and opens the door to challenge.

The main route: demanding the original medium and the chain of custody

Before arguing about artificial intelligence, the classic ground should be exhausted. Case law has long required that, when the authenticity of a digital file is challenged, whoever produces it must prove its origin and integrity. That means identifying the device on which it was generated or received, placing that medium at the court’s disposal and allowing its expert analysis. The defence can request the comparison of the file with the source device, the imaging of the terminal under judicial control and the calculation of the digital fingerprint (hash) verifying that the file has not been altered.

The forensic expert report: what it can and cannot prove

The specific tool against a suspected deepfake is the multimedia forensic expert report. An expert can examine, among other elements, inconsistencies of lighting and shadows, compression artefacts, the coherence between lip movement and the pronounced phoneme, blinking, frame rate, the voice spectrogram and the presence of traces typical of generative models. It is worth being honest about the limits. Automatic detectors of synthetic content have relevant error rates and age quickly, because each new generation of models learns to avoid the traces that betrayed the previous one. That is why a serious report rarely concludes with a categorical “it is fake”: it usually expresses a judgment of probability. And that nuance is precisely what the defence must exploit, because in the criminal process it is not for the accused to prove the falsity of the video, but for the prosecution to prove its authenticity beyond reasonable doubt.

Challenge strategies

The toolbox includes: expressly challenging authenticity, which obliges the prosecution to prove it; requesting the production of the original medium and expert access to it; proposing a party forensic expert report; requesting the testimony of whoever obtained the recording in order to reconstruct its provenance; and, when the material was obtained in violation of fundamental rights, invoking the prohibition of assessment. On the opposite side, whoever produces an authentic recording that is branded synthetic must armour it: preserve the original device, not forward it through applications that recompress it, document the capture date and, if possible, obtain a notarial record or a time-stamped deposit.

Remember: disseminating a manipulated video or audio that falsely attributes conduct to a person can constitute, depending on the case, an offence against honour, against moral integrity or against privacy, and can aggravate liability when used as an instrument of fraud or harassment.

Conclusion

Audiovisual evidence has not died; what changes is the technical rigour with which it must be handled. Challenging a deepfake does not consist in saying the video “looks strange”, but in building a reasonable doubt about its integrity with verifiable arguments. This content is informative and does not constitute legal advice. Every matter requires an individual analysis.