FOLLOW US

What jumped out at me

• A first-of-its-kind moment — a four-minute AI “avatar” of the deceased spoke in a Phoenix courtroom, turning a routine sentencing into a live ethics lab.
• The defense says the video nudged the judge toward a tougher outcome despite no filed aggravators — a potential appellate flashpoint.
• Victims’ rights law is broad onhow a victim can be heard, but AI raises a new question — who gets to put words in the mouth of the dead, and with what safeguards.

By Samuel Lopez — USA Herald

Phoenix, AZ (August 23, 2025) — On May 1, more than twenty people filed into a Maricopa County courtroom for the sentencing of Gabriel Paul Horcasitas, 54, convicted of manslaughter in the 2021 road-rage killing of U.S. Army veteran Christopher Pelkey, 37. As family photos played on a screen, ten speakers gave traditional victim impact statements. Then Pelkey’s sister, Stacey Wales, did something no one in that room had seen before: she pressed play on a nearly four-minute AI-generated video that used Pelkey’s photo and voice profile to deliver words she had written — explicitly introduced as a digital recreation.

The room went quiet. A few people cried. Defense attorney Jason Lamm later admitted his first reaction was, “Is this really happening?” He didn’t object in the moment. He does now believe it lengthened his client’s sentence.

Judge Todd Lang imposed 10½ years for manslaughter — within Arizona’s seven-to-twenty-one-year range — plus a concurrent term for a related endangerment count. Lamm’s frustration isn’t about the existence of victim impact statements; it’s about how this one functioned. Arizona prosecutors hadn’t filed aggravating factors. Under Arizona law, victim comments can be considered, but not as aggravators without proper notice. If the court effectively let the avatar’s sentiments substitute for aggravation, that’s the kernel of the defense’s appellate theory.

“As much as we like to think decisions are purely rational, a lot of sentencing is emotional,” a public defender told me. That’s not cynicism; it’s cognitive science colliding with criminal procedure.

What the law actually says — and doesn’t

Victims have a right to be heard. Arizona’s Constitution (Victims’ Bill of Rights), A.R.S. §13-4442 (impact statements in presentence reports), and the federal Crime Victims’ Rights Act, 18 U.S.C. §3771, all recognize it. None of those provisions micromanage the format. Written, spoken, audio, video — Arizona practice already embraces all of it. On paper, an AI-assisted video is just another container.

But containers carry contents, and contents carry evidentiary risk. Traditional victim allocution can describe harm and humanity. An AI avatar adds performance — facial expressions, intonation, and the persuasive aura of “the victim speaking.” When the words speculate about forgiveness, friendship, or what sentence the deceased might want, we move from memory into mind-reading.

Public defenders’ core objection isn’t to grieving families; it’s to attribution. If the avatar says “I forgive you,” the court can’t test whether the real person would have said that post-homicide. Trauma changes people. That’s not an abstraction; appellate courts routinely caution against punishment decisions fueled by unfair prejudice rather than admissible aggravation.

Wales, to her credit, tried to thread the needle — she avoided having the avatar explicitly forgive or request a number. Still, the format amplifies inference. Even neutral phrasing, when delivered by a lifelike face, can feel like a directive.

Inside the room — how persuasion landed

By multiple accounts, Judge Lang watched intently. The gallery felt it. Wales’ teenage son told her he “needed to hear from Uncle Chris one last time.” None of that is improper; victim healing is a legitimate purpose of allocution. But judges also must police their own minds. That’s the paradox of modern sentencing: the system invites emotion into the room then asks the decision-maker to firewall it from the legal calculus.

As a legal analyst who’s sat through more sentencing hearings than I can count, I’ll tell you straight — everyone thinks they’re immune to persuasive tech. No one is. That’s why procedure exists.

“Give me disclosure, and I can give you due process,” a retired federal judge told me years ago. It still applies, especially here.

The due-process rub the defense is pushing

Arizona requires notice to use statutory aggravators. In this case, the state did not file aggravation. If the court weighed the avatar’s sentiments as functional aggravation — even subconsciously — the remedy could be resentencing. The state will counter that the term was within the presumptive range and supported by mitigators and allocution; the defense will argue the record shows mitigating factors were found but neutralized by improper emotional weight placed on the AI video. Appellate courts don’t often reverse within-range sentences, but novel formats can force a closer look at the line between being heard and being prejudiced.

Where this leaves the Arizona case

Within hours of sentencing, the defense noticed an appeal. Even if the panel finds error, it could call it harmless. But if a remand happens, the opinion will likely read like a policy memo for trial courts nationwide — disclosure, authentication, and a clean record on what the judge did and didn’t rely on. Expect more families to try AI-assisted statements. Don’t expect a flood. Expect new house rules.

“As a former inside-the-system legal analyst, I’m not anti-technology — I’m pro-procedure. The real question isn't whether more families will create AI versions of their deceased loved ones—it's whether our courts will develop coherent standards for evaluating this evidence before the practice becomes routine.” — Samuel Lopez | USA Herald Legal Correspondent