Vancouver police officers thought they had a straightforward way to clean up a social media post. They took a photo of seized drugs and cash from a recent operation. Then they applied software that carried an artificial intelligence label. The result triggered an immediate storm of online criticism.
The original image showed a modest haul. Small amounts of various substances sat next to limited bills on a piece of cardboard marked with Sharpie. Nothing dramatic. Yet the version posted on X carried the tag “made with AI.” Close inspection revealed odd details. Some $50 bills appeared labeled as $20s. A $100 bill displayed “00.” The department quickly pulled the image and replaced it with a cropped original that removed names of the accused.
Sergeant Adam Donalson told CTV News the force used the tool simply to edit out those names. The explanation left many puzzled. The physical labels had been handwritten. Why introduce AI at all? One user replied to the corrected post, “Are you guys claiming this is the ‘real’ photo of what you seized vs the AI slop you posted a few hours ago and deleted?” Another added bluntly, “I like being lied to by the police, it’s good for building trust!”
The episode, reported by Futurism on June 28, 2026, fits a larger pattern. Law enforcement agencies have begun to experiment with generative tools for reports, analysis and visuals. Outcomes have often raised alarms. And the risks appear in plain sight.
Months earlier in Westbrook, Maine, officers faced their own embarrassment. They seized methamphetamine and fentanyl during an arrest on Brackett Street. The evidence photo looked routine enough. But when an officer wanted to add the department logo for a Facebook announcement, he turned to ChatGPT. The app did more than insert a badge. It changed packaging, altered lettering on bags, removed visible residue and modified other objects including a spoon and case.
Capt. Steven Goldberg described the outcome as bizarre. “When we look and compare the two photos, it’s really bizarre as to what that app did to the photo,” he told WGME. “It got rid of those cookie cutter packaging, it completely redid the lettering of some of the packages and it seemed to alter some of the things we took pictures of. I know we mentioned the spoon and the case, and we don’t have a good explanation why.” The department apologized, invited media to view the actual evidence and insisted the seizure remained legitimate. Charges proceeded. Yet the damage to public confidence lingered. Resident Jessica Wellman observed, “The fact that the person who posted it and put it through ChatGPT didn’t notice the differences because they were very obvious…it makes me wonder how much people understand about technology and how easy it is to fool people.”
These incidents stand out because they involve visual evidence shown directly to the public. But the problem runs deeper. Prosecutors have started to draw lines. In September 2024 the King County Prosecuting Attorney’s Office in Seattle announced it would no longer accept any police reports generated with artificial intelligence assistance. Narratives must come entirely from the authoring officer. The Fair and Just Prosecution brief highlighted the absence of oversight. No one tracks exactly how many departments rely on such tools. Accuracy suffers. Hallucinations creep in.
California responded with legislation. San Diego Police Department issued a December 2025 order barring officers from using generative AI in reports unless specifically approved and covered by training policy. The move aligned with a new state law requiring disclosure whenever AI assists in drafting or compiling documents. Departments elsewhere show mixed approaches. Some see promise in processing wiretap transcripts, mapping supply chains or spotting patterns in narcotics cases. Others worry about admissibility in court.
But the Vancouver and Westbrook cases hit a nerve precisely because they involve images presented as evidence of success. One netizen warned under the Vancouver replacement photo that jurors might see the original AI version. “You showed an AI-generated image of fabricated evidence to the public, which includes potential jurors. This whole case is going to get thrown out.” Whether that fear materializes remains uncertain. The point is clear. Once trust erodes, every future claim faces extra scrutiny.
Industry publications and researchers have tracked the trend. Police1 described how generative systems can accelerate analysis of gang relationships, fentanyl networks and fraud cases. Speed gains are real. Yet the same article stressed the necessity of human oversight and ethical guardrails. Without them, errors compound quickly. A UK example from West Midlands Police showed Microsoft Copilot inserting a fictitious football match into an intelligence report. The hallucination influenced decisions on fan attendance. An investigation followed.
So what drives departments toward these tools? Volume of data continues to grow. Officers face pressure to produce reports faster. Vendors promise efficiency in sifting evidence, summarizing documents and generating leads. The Department of Justice’s December 2024 report on AI in criminal justice acknowledged expanding applications while flagging persistent operational and civil rights challenges. It noted machine learning already helps classify drug origins and analyze evidence. Generative systems add another layer. They also multiply the chance of undetectable mistakes.
Public reaction on X reflected broad skepticism. Posts following the Vancouver incident questioned whether agencies might fabricate bodycam footage or court exhibits next. One user tied the issue to declining faith in institutions. Another pointed to similar experiments abroad, including Thai police generating playful AI images after real raids. The common thread? Once altered content reaches the public, distinguishing real from synthetic grows harder. And police shoulder the burden of proof.
Experts caution against outright bans. Properly governed AI can expose supply chain choke points in narcotics trafficking or highlight recruitment patterns faster than manual review. The Europol report on AI and policing outlined capabilities from advanced analytics to biometric identification. It also mapped regulatory frameworks designed to keep systems accountable. The question is whether individual departments possess the technical literacy to deploy these capabilities without stumbling.
In Westbrook the captain admitted the department lacked a good explanation for the changes. That candor helped contain immediate fallout. They showed the physical evidence to reporters. The investigation moved forward. Yet the episode illustrates a broader gap. Training on generative tools remains inconsistent. Many officers treat them like simple photo editors. The models operate on probabilities, not truth. They fill gaps with plausible inventions. When those inventions appear in evidence photos or reports, the consequences stretch beyond embarrassment.
Similar concerns surfaced in the UK where Derbyshire Police placed an officer on non-frontline duties after allegations of using AI to create evidence. A criminal investigation examined potential perversion of justice. The case remains active. Each new story adds to the cumulative doubt.
Agencies that move fastest sometimes pay the highest price in credibility. Vancouver replaced its image within hours. Westbrook issued an apology and opened its evidence room. Both responses show awareness of the optics. Neither fully resolves the underlying tension. Generative AI is already inside police workflows. The public now sees the output. And the margin for error has narrowed to zero.
Future policy may demand watermarking, audit logs and mandatory disclosure for any AI-assisted visual or narrative evidence. Some prosecutors already reject AI-generated content outright. Others may follow. Technology vendors will likely adapt by building law-enforcement-specific models with stricter factual anchors. Until then, departments walk a fine line. They gain efficiency. They risk something more valuable. The appearance of integrity matters as much as the reality when juries and citizens decide whom to believe.
The Vancouver post carried a modest seizure. The Maine one involved serious substances. Neither case collapsed because of the AI edits. But both revealed how quickly a single image can undermine years of community relations. Police work has always depended on trust. Generative tools test that foundation in new and unpredictable ways. The coming years will show whether departments can master the technology before the technology masters the narrative.
WebProNews is an iEntry Publication