When AI Speaks for Someone Who Can't: A Court Just Drew a Line
A Havoc Override Briefing
Artificial intelligence can recreate a face, a voice, and words someone never actually said. A recent Arizona court decision shows why that distinction can matter enormously once synthetic content enters a courtroom.
On September 30, 2026, the Arizona Court of Appeals vacated the manslaughter sentence of Gabriel Horcasitas and ordered resentencing after the sentencing court considered an AI-generated video depicting deceased victim Christopher Pelkey speaking directly to the court and the defendant. The conviction itself remained intact. The appellate court concluded that consideration of the AI-generated portion of the video rendered the sentencing fundamentally unfair.
What people think is happening
It is easy to think of AI-generated material as simply another kind of visual aid.
A recreated voice. A reconstructed image. A simulation intended to help a court understand something that cannot otherwise be shown.
But synthetic material creates a fundamentally different question from an ordinary photograph, recording, screenshot, or video:
What part of this actually happened?
What's actually happening
The problem wasn't simply that AI was used
The Arizona decision did not establish a blanket rule that every use of artificial intelligence in court is prohibited.
The appellate court distinguished between genuine footage embedded in the presentation and the AI-generated depiction of the victim. The problem was that the synthetic portion presented words, expressions, and sentiments as though they came directly from a person who never actually made them.
That distinction matters.
There is a significant difference between using technology to organize, clarify, enlarge, transcribe, or illustrate existing information and using technology to create new purported speech, conduct, or evidence that never existed in the underlying record.
Synthetic realism creates a credibility problem
AI-generated video can look and sound increasingly convincing.
That visual realism can create emotional or persuasive force beyond what the underlying information supports.
In the Arizona case, the appellate court specifically noted that the sentencing judge described the AI depiction as genuine and referenced the apparent forgiveness expressed by the simulated victim. The appeals court concluded there was no question the video played a role in sentencing.
This is precisely why provenance is becoming more important.
Where did a file come from?
Who created it?
Was it altered?
Was AI used?
Does an original exist?
Can the underlying source material be produced?
Those questions are becoming part of modern evidence literacy.
The problem works in both directions
AI does not only make fabricated material easier to create.
It can also make authentic material easier to challenge.
As courts become more aware of deepfakes, synthetic audio, altered screenshots, and generated documents, someone presented with genuine digital evidence may increasingly hear a new response:
“That's fake.”
The National Center for State Courts has already identified this emerging problem and has issued guidance for courts dealing with acknowledged and disputed AI-generated evidence.
That makes preservation more important, not less.
Original files, metadata, source devices, complete message threads, download histories, timestamps, and corroborating records can help establish where digital material actually came from.
Why this matters strategically
For anyone documenting a high-conflict situation, the lesson is not “never use AI.”
The lesson is to maintain a clear boundary between what actually occurred and anything technology later creates, enhances, summarizes, or illustrates.
If an original recording exists, preserve the original.
If screenshots are taken, preserve the underlying conversation where possible.
If AI is used to transcribe, organize, summarize, enhance, or visualize information, maintain the source material separately.
And never assume something becomes stronger evidence simply because technology makes it more persuasive.
What knowing this changes
You begin thinking about digital material in terms of provenance, not merely appearance.
A compelling image is not automatically reliable.
A realistic voice is not automatically authentic.
A polished reconstruction is not the same thing as a contemporaneous record.
As synthetic media becomes easier to create, the boring things become increasingly valuable:
Originals.
Metadata.
Dates.
Sources.
Context.
Chain of custody.
Technology changes quickly.
The need to establish what actually happened does not.
Module 2 goes further.
Motion & Filing Guide: How To Navigate Court goes deeper into court procedure, evidence preparation, hearings, filings, and strategic preparation for representing yourself in a legal proceeding.
Trying to organize evidence or prepare for an active legal proceeding right now?
The 1:1 Strategic Analysis maps your circumstances and provides an individual written strategy.
Start Your 1:1 Strategic Analysis
Havoc Override provides strategic self-advocacy education. This is not legal advice. Rules governing evidence, admissibility, authentication, and the use of AI-generated material vary by jurisdiction and proceeding. Consult a licensed attorney in the applicable jurisdiction for legal advice specific to your situation.