Deepfakes in the courtroom, and the race to prove what is real

Courts are confronting a problem that would have sounded like science fiction a few years ago: evidence that may have been generated by AI. US news reports this year describe judges and lawyers struggling with how to authenticate video, audio and images when convincing fakes are easy to make. A report by Sinclair's Spotlight on America described experts warning that safeguards lag behind the technology. A Mexican outlet, N+, reported that OpenAI and Meta had warned the US judiciary that fake videos are becoming hard to detect for people and software alike.

At the same time, researchers keep building better detectors. ScienceDaily reported on a UCLA research prototype that uses light to screen around fifteen videos at once, detecting deepfakes with nearly 98% accuracy on a benchmark. Coverage noted that accuracy on videos from a newer generator came after extra fine-tuning, a reminder that detectors must keep up with each new model.

The two-sided problem

Deepfakes create two problems for courts, and for everyone else:

1. Fake evidence may be presented as real.
2. Real evidence may be dismissed as fake. Lawyers have begun arguing that genuine video "could have been" generated, a tactic sometimes called the "liar's dividend".

Both erode trust. If anything can be faked, everything can be doubted.

Detection alone will not solve it

Detectors improve, and so do generators. Many experts argue that the more durable answer is provenance: knowing where a piece of media came from, who made it and whether it has been altered. That means verified chains of custody for evidence, cameras and apps that sign their recordings, and platforms that show where content originated.

Why Kenya should pay attention

Kenya's institutions will face the same questions. The Independent Electoral and Boundaries Commission has listed AI-driven misinformation and deepfakes among the risks to the 2027 general election, and the Media Council of Kenya's election coverage guidelines ask outlets to detect and correct synthetic media. AI and the 2027 election: a code of conduct for brands.

Courts, banks, employers and newsrooms will all need ways to check what is real.

What brands and creators can do

AI video is a powerful creative tool, and Tari Studio is one. Being part of the solution rather than the problem is a choice every user makes:

• Never imitate a real person without their permission, and never put words in a real person's mouth.
• Do not depict real events that did not happen, especially news, crimes or accidents.
• Label AI-generated content where audiences could be misled. When to label AI content.
• Keep your records: the prompt you used, the files you attached, the date.

In Tari Studio, every result keeps the Prompt we sent, so you can show exactly what the model was asked to make. Honest creators will increasingly need to prove their work is what they say it is. Good records make that easy.