Editorial: When seeing is no longer believing

AI-created videos may pollute attention spans with energy-guzzling digital banalities but, inside the Maltese courts, the threat of deepfakes could be set to become one of the most serious threats to judicial integrity.

Hyper-realistic AI-generated video, audio and imagery can convincingly depict events that never occurred, or conjure up statements that were never uttered. When the entire world’s justice systems have been diligently built on the assumption that audiovisual evidence is inherently trustworthy, then it can safely be said that, in 2026, we could be entering a reality where seeing is no longer believing.

It’s clear that social media giants are unwilling to ward off AI slop: content generated by cheap, low-cost apps gives a steroidal bump to advertising revenue derived through addictive short-video churn.

Instagram CEO Adam Mosseri conceded that AI slop was taking over his platform but gingerly suggested that users should “keep track of real human-generated content” so that their algorithms return them with real content.

That declaration showed an unwillingness to control AI content. But throwing up hands in despair, for fear of losing some money, is hardly acceptable.

For social media users alone, a world in which we have to move from assuming what we see is real by default to starting with scepticism is a horrific prospect. Humans are inherently predisposed to believing their eyes. So what happens when this threat arrives at the doorstep of our courtrooms?

If a witness testifies a video is accurate or if a court expert testifies that metadata looks correct, courts will consider that sufficient. Deepfakes will shatter this trust if videos and audio can be artificially generated and human observation cannot reliably distinguish fake from real.

The government has rightly responded with a pre-emptive strike against this threat, realising that, without immediate action, the Maltese courts could face the prospect of evidence being undermined by doubts cast over their veracity in a time where AI turns truth into a fluid quantity.

If deepfake technology evolves faster than forensic detection, the court experts who we depend on to verify metadata, device records and file integrity, might find it hard to establish who created manipulated content.

One of the strongest recommendations against abusive deepfakes in courts must be procedural: in compilations of evidence, digital evidence legitimacy should undergo a high standard of gatekeeping, where magistrates ensure only credible material reaches judges and juries.

Expert testimony will remain indispensable in evaluating metadata integrity and manipulation traces, but also to explain findings in understandable language to judges who might not be digitally literate. This latter point is crucial as well: judges must undergo training programmes on AI and digital manipulation and be able to understand deepfakes conceptually.

Courts, the police and the office of the attorney general should also invest in vetted AI forensic tools that can provide the capability to detect manipulated evidence.

Humans naturally trust high-quality visual evidence – perhaps we have been careless of the effect this has had on social media users too old to be digitally literate or too young to understand what content they are being fed.

Clearly, the psychological effect of seeing or hearing something is stronger than abstract testimony. So if deepfakes start challenging truth itself, courts and juries will likely be unable to discern falsehoods, and the ground beneath us will shift – even unadulterated reality will be questioned.

This is a wake-up call for our decision-makers. We can work now to preserve the credibility of justice or sleepwalk into a crisis of public trust.