AI-crafted Victim Statements Risk Compromising Justice Integrity

When a police officer in Nottingham used Microsoft Copilot to draft a victim impact statement with instructions to “make a judge or reader weep” and reach “the highest sentence,” the case exposed a troubling frontier where artificial intelligence meets criminal justice.

Jerome Gibson, 23, received a 44-month jail sentence after threatening his partner with a knife and injuring a police officer. His sentence appeal was rejected by the Court of Appeal, but not before revealing that the victim statement submitted in his case had been AI-generated by a police officer who explicitly told the software to maximize emotional impact and secure the harshest possible punishment. The officer has since been removed from front-line duties. While the court determined that the AI-generated statement didn’t actually extend Gibson’s sentence, the incident raises fundamental questions about fairness, authenticity, and the role of technology in shaping criminal outcomes.

Victim impact statements serve a specific purpose in sentencing: they give victims a voice in court and help judges understand the real-world consequences of criminal acts. These statements are meant to reflect genuine experiences, emotions, and harm. When a police officer feeds prompts into an AI system asking it to create language that will “make a judge or reader weep,” that purpose gets corrupted. The statement becomes a performance engineered for maximum emotional effect rather than an authentic account.

Lucky Thandi, representing Gibson in the appeal, noted that the officer “had made great efforts to ensure that the victim personal statement had read as powerfully as it possibly could.” That phrasing suggests intent. The officer wasn’t using AI to help a victim articulate existing feelings. The goal was manufacturing emotional intensity that might not reflect what the victim actually experienced or wanted to convey.

This kind of emotional manipulation matters because judges are human. They’re supposed to weigh evidence objectively, but victim statements influence sentencing decisions. Research has consistently shown that powerful victim impact statements correlate with harsher sentences. When that power comes from algorithmic wordsmithing rather than genuine testimony, defendants face punishment calibrated not to the actual harm they caused but to how effectively an AI tool can tug at judicial heartstrings.

The Gibson appeal exposed the dangers of using generative AI to craft victim impact statements for maximum emotional and punitive effect. Although the Court of Appeal found that the statement did not lengthen Gibson’s sentence, the episode highlights broader risks: fabricated or distorted facts, hidden AI involvement, weakened rights to challenge evidence, and biased or manipulative uses of technology in policing. Mandatory disclosure, independent auditing, clear ethical rules, training, and meaningful enforcement are needed before AI becomes embedded in criminal justice documentation.

Beyond emotional manipulation, AI-generated police documentation carries significant accuracy risks. Multiple studies and pilot programs examining AI police report writing have documented serious problems. These systems have been shown to fabricate dialogue that never occurred, misidentify individuals, include officers who weren’t present at scenes, and insert details that contradict body camera footage.

These aren’t minor typos. When an AI system hallucinates facts or invents statements, those fabrications enter the official record. Prosecutors rely on police reports and victim statements to build cases. Defense attorneys use them to prepare responses. Judges reference them during sentencing. If the foundational documents contain AI-generated inaccuracies, every subsequent decision built on that foundation becomes suspect.

The Gibson case involved a victim statement presented as though written by the victim herself (referred to as F in court documents). But F didn’t write it. An officer used AI to compose it, apparently without adequate disclosure that the statement was computer-generated. That’s a transparency failure on top of the authenticity problem. Criminal defendants have a right to understand the evidence against them and to challenge its accuracy and source. When AI-generated content gets presented as human testimony without clear labeling, that right erodes.

The Nottingham case doesn’t exist in isolation. Police departments across multiple jurisdictions have started experimenting with AI for report writing, evidence analysis, and documentation. Some agencies promote these tools as efficiency measures that free officers from paperwork and let them spend more time on patrol. Others highlight AI’s potential to standardize reporting quality and reduce human error.

Those benefits might sound appealing, but they come with substantial risks that many departments haven’t adequately addressed. Research from the Council on Criminal Justice examining AI report-writing tools found mixed evidence on time savings and raised concerns about reliability when AI systems analyze body camera footage to generate incident narratives. The Policing Project has documented that few jurisdictions require police to disclose when they’ve used AI in investigations or reports, creating a transparency vacuum that hampers oversight and accountability.

The Brennan Center for Justice’s research on AI in policing emphasizes how these systems often amplify existing biases in law enforcement. AI tools trained on historical arrest records and crime data inherit the racial and socioeconomic disparities baked into that data. When police use AI-driven data fusion tools to predict crimes or identify threats, communities of color face disproportionate targeting. The same bias problems that plague predictive policing algorithms can infect AI-generated documentation if the training data or prompts reflect skewed assumptions.

What makes the Gibson case particularly alarming is the explicit instruction to maximize punishment. The officer didn’t just use AI as a drafting aid. The prompts reveal an agenda: reach the higher harm category, achieve the highest sentence, make the reader weep. That crosses a line from assistance to advocacy, turning a supposedly neutral documentation tool into a weapon for securing harsher outcomes.

AI may help organize information or reduce administrative burdens, but it should never be used to manufacture a victim’s voice or optimize punishment. In a criminal case, every document carries consequences, so the standard must be higher than whether an AI-generated statement sounds persuasive or whether one appeal produces no measurable change. Authenticity, traceability, verification, and the ability to challenge evidence should be treated as nonnegotiable safeguards.

The legal profession has started grappling with ethical obligations around AI use, but the rules haven’t caught up to the technology’s rapid adoption in criminal justice contexts. The American Bar Association’s Model Rules require lawyers to maintain competence, verify AI-generated work, and uphold candor to the court. Multiple states have issued guidance emphasizing that lawyers can’t let AI substitute for their professional judgment and must supervise AI use carefully.

Those standards apply to attorneys, but police officers operate under different rules and oversight structures. Law enforcement agencies don’t universally have policies governing AI use in evidence gathering or documentation. Training on AI risks and limitations remains inconsistent. Without clear guidelines, individual officers make judgment calls about when and how to deploy AI tools, sometimes with results like the Gibson case.

Denis Barry, representing the Crown Prosecution Service in the Gibson appeal, acknowledged the problem: “The conduct of the officer is concerning, not so much because of the use of technology, but it displays a use of evidence gathering that is concerning.” That framing suggests the issue extends beyond AI specifically to how officers approach their documentation responsibilities. If an officer thinks it’s acceptable to engineer victim statements for maximum punitive impact, that’s a problem whether they use AI or not. But AI makes it easier, faster, and more sophisticated, amplifying the potential for abuse.

Federal Rule of Civil Procedure 11 imposes a nondelegable duty on lawyers to validate the accuracy of court filings. Similar principles should apply to police officers submitting evidence and statements in criminal cases. The duty to verify shouldn’t disappear just because an AI system generated the content. Yet without explicit policies and enforcement mechanisms, that verification often doesn’t happen.

Several advocacy organizations and legal experts have proposed mandatory disclosure requirements as a starting point for addressing AI use in criminal justice. The Policing Project has drafted model legislation requiring police agencies to disclose when they’ve used AI in reports and to maintain publicly available policies on AI use. Such transparency would enable prosecutors, defense attorneys, and judges to evaluate how AI influenced investigations and documentation.

Disclosure alone doesn’t solve the accuracy or bias problems, but it creates accountability. If a defense attorney knows a victim statement was AI-generated, they can question its authenticity and probe whether it accurately reflects the victim’s actual experience. If judges understand that AI was used, they can apply appropriate skepticism and request corroborating evidence. If the public can access police AI use policies, community oversight becomes possible.

The absence of transparency creates the opposite dynamic. When AI use goes undisclosed, everyone in the courtroom operates on false assumptions. They treat AI-generated content as authentic human testimony. They don’t know to look for hallucinations or fabricated details. The defendant’s right to confront evidence and challenge its reliability gets undermined.

The Court of Appeal’s decision to reject Gibson’s sentence appeal sends a mixed message. On one hand, the court called the officer’s conduct “deplorable,” acknowledging the serious ethical breach involved. On the other hand, the court determined that the AI-generated statement didn’t actually make Gibson’s sentence longer than it would have been otherwise.

That second part is troubling because it treats the AI use as a harmless error. The logic seems to be: yes, what the officer did was wrong, but Gibson deserved the sentence anyway based on his actions, so no harm, no foul. That reasoning might work for this specific case, but it fails to address the systemic implications.

How can anyone confidently assert that the AI-generated statement didn’t influence the sentence? Judges don’t announce alternative sentences they would have imposed under different evidentiary circumstances. The court in Gibson’s case presumably reviewed multiple statements and documents, and the judge brought experience to the sentencing decision. But acknowledging that doesn’t prove the AI-generated statement had zero impact. Victim statements are designed to influence sentencing. That’s their entire purpose. An AI-crafted statement optimized for emotional impact likely did what it was designed to do.

Moreover, the “no harm” conclusion in Gibson’s case doesn’t help future defendants who might face similar AI manipulation under circumstances where the impact can’t be dismissed so easily. The ruling doesn’t establish clear guardrails or consequences that would prevent other officers from using AI the same way.

Criminal sentencing determines how long someone stays incarcerated, how their family copes with their absence, what opportunities they forfeit, and how they reintegrate into society afterward. These are high-stakes decisions affecting real people’s lives. When AI enters that process in ways that distort evidence or manufacture emotional appeals, ordinary people bear the consequences.

Gibson committed serious offenses. His appeal was rejected, and the court determined his sentence was appropriate. That doesn’t erase the problem with how the case was handled. Criminal defendants deserve fair proceedings regardless of their guilt. The process matters, not just the outcome. When police officers use AI to engineer victim statements for maximum punitive effect, that fairness breaks down.

Victims deserve authentic representation too. If F, Gibson’s partner, had specific things she wanted to communicate to the court about how the assault affected her, those genuine sentiments should have been what the judge heard. Instead, she got a statement crafted by an algorithm following prompts designed to maximize her boyfriend’s sentence. That’s not empowerment. That’s ventriloquism.

The Gibson case should serve as a warning. AI tools will keep getting more sophisticated and more widely adopted in policing and criminal justice. Without proactive governance, the problems evident in Nottingham will multiply. Several steps would help:

Mandatory disclosure policies requiring police to identify when and how they used AI in evidence gathering and documentation. These policies should apply across all phases of criminal cases, from initial reports through victim statements and testimony preparation.

Independent oversight and auditing of police AI tools to assess accuracy, bias, and reliability. Vendors should be required to submit their systems for testing before deployment, and ongoing monitoring should catch problems that emerge in real-world use.

Clear ethical guidelines for police officers on appropriate and inappropriate AI use. These guidelines should emphasize that AI can’t be used to manipulate evidence, manufacture emotional appeals, or undermine the authenticity of victim and witness statements.

Training for all criminal justice system participants on AI capabilities, limitations, and risks. Judges, prosecutors, and defense attorneys need to understand how AI-generated content can differ from human testimony and what questions to ask when AI use is disclosed.

Enforcement mechanisms with real consequences when officers misuse AI tools. Removing someone from front-line duties after the fact, as happened in the Gibson case, doesn’t address whether the conduct will trigger disciplinary action or deter similar behavior by other officers.

None of these measures require halting AI use entirely. The technology has potential legitimate applications in policing and criminal justice. But that potential needs to be balanced against the substantial risks to fairness, accuracy, and defendants’ rights. The current approach, where individual officers experiment with AI tools guided mainly by their own judgment and instructions to make judges weep, clearly doesn’t provide that balance. The people most affected by these decisions, defendants and victims alike, deserve better.