Arizona court rejects AI victim video, fueling demand for deepfake detection

An Arizona appeals court has vacated the 10.5-year sentence of Gabriel Horcasitas, who was convicted of manslaughter in a 2021 road-rage shooting, after ruling that the sentencing judge should not have considered an AI-generated video of victim Christopher Pelkey.

The conviction still stands, but the case will return for resentencing, according to Reuters and a case summary.

A forgiveness message the panel said tainted the sentence

The three-judge panel said the AI-generated video improperly influenced the sentencing.

While the record does not reflect precisely how the AI video factored into the sentencing calculus, there is no question it played a role.

— Judge D. Steven Williams, Arizona Court of Appeals, as reported by Reuters

The court concluded that the video prejudiced Horcasitas enough to make the sentencing procedure fundamentally unfair. Reuters reported that Arizona’s attorney general and Horcasitas’s public defender declined to comment.

How Pelkey’s family built the avatar

The script was written by Pelkey’s sister, Stacey Wales, who found it difficult to express in words what she thought her brother would have said in that situation. According to NPR, the family created the avatar by means of a short video clip, an image from the funeral, and a variety of other AI tools.

The avatar introduced itself as an AI creation before speaking to Horcasitas with the words:

In another life, we probably could have been friends.

Later, the judge of the trial expressed gratitude to the family for the video.

Courts have no shared rulebook for synthetic video

The decision comes at a time when courts face challenges when it comes to artificial evidence. A report published by the University of Colorado Boulder states that over 80% of court cases in the U.S. are partly based on video evidence, while courts are yet to have common rules to deal with AI-generated or AI-enhanced footage.

Researchers are also concerned about the so-called “deepfake defense,” whereby real video footage is declared as fake because AI makes the claims plausible.

This worry was one of the reasons for the creation of CIFAR Synthetic Evidence Corpus, initiated in June 2026, as researchers realized that previous datasets were unsuitable for teaching the technology how to detect subtle manipulations that could occur in evidence material.

Provenance, watermarks, and a detection market racing to catch up

NIST identifies provenance tracking, watermarking, and synthetic-content detection as key ways to manage AI-generated media. Europe has gone further: Article 50 of the EU AI Act requires certain synthetic content to be machine-readable and deepfakes to be disclosed as artificially generated or manipulated.

The commercial link is becoming easier to see. SNS Insider values the deepfake-detection market at $1.19 billion in 2026 and projects it will reach $12.14 billion by 2035, a 29.5% CAGR.

One Arizona ruling will not create that market by itself. But when more than four-fifths of court cases already depend on video, every dispute over whether evidence is authentic increases the practical need for tools that can verify it.

That turns courtroom deepfakes from a legal headache into a real business opportunity for companies selling forensic verification. China is moving in the same direction, with Cryptopolitan reporting new liability rules for AI deepfakes and voice cloning in September.