Disadvantages of Facial Recognition: What Killinger's Case Proves for Investigators

A UPS driver in Reno showed officers his Nevada driver's license. He showed them his pay stub. He showed them his vehicle registration. The casino's facial recognition system had already decided none of that mattered, it said he was a trespasser, and it said so with 100% confidence. He spent 11 hours in custody anyway.

That case, now heading toward a 2026 trial, is not a technology story. It is a validation story. And if you investigate people for a living, it should be keeping you up at night, because the underlying logic failure that put Jason Killinger in handcuffs is the same one quietly sitting inside thousands of ongoing investigations right now.

Facial Recognition False Arrest: The 100% Confidence Problem

The Peppermill Casino in Reno had a trespasser on file. Their system flagged Killinger as a match. On paper, that sounds like due diligence. In practice, Casino.org reports that the arresting officer has since admitted under oath that the arrest "never should have happened", and the lawsuit alleges he knowingly inserted false statements into police reports claiming Killinger's legitimate ID documents were fraudulent.

There was also, apparently, a four-inch height difference and mismatching eye color between Killinger and the actual trespasser. You'd think those details might give someone pause. They did not.

"Facial recognition should be treated as an investigative lead only, requiring further corroboration before arrest." Arresting officer, under deposition oath, as reported by State of Surveillance

Here's the problem: that's not what happened. The score was high, the system said "match," and everything else, physical documentation, observable physical differences, basic common sense, got subordinated to an algorithm's confidence rating. That's not the algorithm failing. That's the human workflow failing, and it's a distinction that will matter enormously when this goes to trial.

The full case timeline, including the constitutional violation allegations, is documented by All About Lawyer. What it describes is a cascade: a machine produces a number, a human interprets that number as certainty, the system around that human has no protocol for pushback, and an innocent person pays the price. Replace "casino security" with "private investigator submitting evidence in a civil matter" and the logic holds exactly the same way.

Facial Recognition Evidence Chain: When Systems Fail

What makes the Killinger case genuinely alarming isn't that it happened. Biometric false positives have always existed. What's alarming is the context it landed in, the same week, same news cycle as a wave of stories that all say the same thing from different angles: you can no longer treat visual or audio evidence as self-authenticating.

Elderly people across multiple countries are being conned by AI-generated voices and video impersonating government officials. Deepfake pornography is spreading fast enough that the EU is scrambling for a legislative response. In Australia, deepfake videos of a sitting Premier are circulating on social media ahead of elections, prompting warnings from media commentators about AI's capacity to manufacture political reality. Meanwhile, according to Ballotpedia News, 15 deepfake-specific bills have been enacted in the United States in the current legislative year alone, which tells you exactly how fast this moved from "tech curiosity" to "actual legal emergency."

The elder fraud angle deserves its own moment of attention. According to Journal of Accountancy, AI-powered scams targeting seniors, voice cloning, deepfake video, sophisticated phishing, contributed to $4.89 billion in total elder fraud losses in 2024, with an average loss of $1,298 per incident. These aren't abstract statistics. These are investigators' future clients, future cases, and future witnesses whose credibility will be challenged the moment opposing counsel points out they were tricked by a synthetic voice they genuinely couldn't distinguish from a real one.

Evidence Chain Protocols: A Stress Test for Investigators

Here's the uncomfortable question this week demands: when you receive a key photo or video in a case today, what is your actual validation process? Not the one you'd describe in a deposition. The actual one.

If the honest answer is "I look at it, it seems legitimate, I move forward", that process is now a liability. Not because it was ever particularly rigorous, but because the bar for what opposing counsel can challenge has just risen dramatically. A year ago, raising deepfakes in court was a fringe defense move. Today, with 15 new state laws acknowledging that synthetic media is a genuine legal threat, a skilled attorney asking "how did you verify this image wasn't AI-generated?" is not a Hail Mary. It's a standard cross-examination question you should be ready to answer with documentation, not hope.

The Vectra AI research on AI scam detection frames this neatly as "truth decay", the gradual erosion of trust in any digital interaction because the cost of fabricating convincing fakes has dropped to near zero. That's not hyperbole. That's the environment every investigator is now working in, whether they've acknowledged it yet or not. And the investigators who haven't updated their mental model are the ones most exposed.

The fix, to be clear, is not to stop using visual evidence. Photographs and video remain among the most powerful evidentiary tools available. The fix is to treat them as claims that require verification rather than facts that speak for themselves, and to document that verification process in a way that survives aggressive cross-examination. That means provenance tracking, metadata analysis, source chain documentation, and where facial comparison is involved, understanding exactly what the tool's false positive rate is in the specific context where it was used. A match rate that performs well on a curated test dataset may perform very differently in real-world casino lighting conditions. Knowing that distinction matters when your work is the last line between a flawed "100% match" and someone else's 11 hours in a cell.

Accuracy: Why Facial Recognition Still Gets It Wrong

Accuracy is the most obvious of the disadvantages of facial recognition, and the Killinger case is a plain demonstration of it. A facial recognition system can be less accurate than its stated confidence score suggests, especially once you move away from clean, well-lit enrollment photos into messy real-world conditions like casino floor lighting, odd camera angles, or a partial view of someone's face. Facial recognition tends to perform worse on exactly the kind of grainy, off-angle footage that investigators actually work with, which means a "100% match" label can mask a much shakier underlying result. Treating that number as gospel, rather than as one data point among several, is how accuracy problems turn into wrongful arrests.

Facial Spoofing and Recognition Challenges

Facial spoofing, using a photo, mask, or manipulated image to trick a facial recognition system into a false match, is another one of the recognition challenges investigators need to understand. As synthetic media tools get better and cheaper, the line between a genuine face and a convincingly faked one gets thinner, and facial recognition systems trained on older data may not catch the difference. This is exactly why a high-confidence facial recognition result should be treated as a lead to confirm, not a conclusion to act on. Anyone building an evidence chain around facial recognition output needs a plan for what happens when the underlying image itself can't be fully trusted.

Privacy Concerns and Data Security

Privacy concerns sit right alongside accuracy problems on any honest list of facial recognition drawbacks. Facial data and other biometric data are uniquely sensitive because, unlike a password, you can't simply change your face if a database is breached, so data security failures around facial recognition systems carry consequences that last far longer than a typical data leak. There's also the matter of user consent, many people photographed by casino security cameras, retail access control systems, or public cameras never agreed to have their facial data compared against a watchlist. That gap between "your face was scanned" and "you agreed to be scanned" is at the center of ongoing privacy debates about facial recognition technology.

Bias, False Negatives, and Public Perception

Bias in facial recognition systems, where accuracy rates differ across skin tones, ages, or genders, has been documented for years and remains one of the sharper disadvantages of facial recognition in practice. That bias shows up as both false positives, like the one that put Killinger in handcuffs, and false negatives, where a legitimate match is missed entirely, letting a real security risk walk right past an access control checkpoint. Public perception of facial recognition technology has shifted as a result, with more people asking hard questions about where their face recognition data goes and who reviews it. It's hard to avoid facial recognition entirely in daily life now, from unlocking a phone to walking through an airport, which is exactly why understanding these risks matters for anyone whose work touches this technology.

None of this means facial recognition technology is useless, it remains a genuinely useful tool for narrowing down leads quickly. But the risks it poses, from accuracy gaps to privacy concerns to outright facial spoofing, mean facial recognition output needs the same corroboration standard as any other single piece of evidence. The Killinger case shows what happens when that standard gets skipped: a confident-sounding number replaces the judgment that a trained human is supposed to apply. For investigators, the lesson isn't to abandon facial recognition technology, but to build a workflow where its output is always the start of the inquiry, never the end of it.

Real-World Conditions and Why Lab Accuracy Doesn't Transfer

Facial recognition vendors love to quote accuracy numbers pulled from controlled lab testing, but real-world conditions rarely match a lab. Casino floors, parking lots, and airport corridors bring inconsistent lighting, motion blur, and partial face coverage that a laboratory benchmark never accounts for. A facial recognition system rated at near-perfect accuracy in testing can degrade sharply once it's pointed at a grainy security camera feed in real-world conditions. That gap between advertised accuracy and street-level performance is exactly the gap that swallowed Jason Killinger.

Investigators relying on facial recognition technology should ask vendors pointed questions about how accuracy was measured, and under what conditions. A recognition system tested only on high-resolution, front-facing enrollment photos tells you very little about how it performs on a blurry casino camera angle. Demanding real-world accuracy data, not marketing-sheet accuracy data, is a small step that can prevent a large mistake.

It's Hard to Avoid Facial Recognition in Daily Life

It's hard to avoid facial recognition today even for people who actively try. Airports use it for boarding, phones use it to unlock, retailers use it for loss prevention, and casinos use it for security, often without a clear, visible opt-out. This creeping default use is part of why public perception of facial recognition technology has soured even among people who have never personally been misidentified.

For investigators, this ambient presence of facial recognition cuts both ways. It means more available footage and more potential leads, but it also means more raw material that could contain a false match, a spoofed image, or a low-confidence result mislabeled as certain. Because it's hard to avoid facial recognition technology entirely, the responsible move is building a verification habit around it rather than hoping it simply gets more accurate on its own.

Data collection is the quiet engine behind every facial recognition system in use today, and it deserves more scrutiny than it usually gets. Every scan, whether at a casino entrance or an airport gate, adds another record to a growing pool of biometric data that someone, somewhere, is responsible for securing. When that data collection pipeline has weak controls, the downstream risk isn't just a bad match, it's a permanent, unchangeable identifier sitting in a database that could eventually leak. Investigators who rely on facial recognition output should ask where the underlying data collection happened, how long it's retained, and who has access control over it.

Facial recognition technology raises privacy concerns that go well beyond a single wrongful arrest. Recognition technology built for security purposes often gets repurposed for marketing, employee monitoring, or law enforcement lookups far removed from its original justification, and few people photographed by a security camera ever consented to any of those downstream uses. This kind of scope creep is exactly why privacy concerns around facial recognition keep growing louder even as the underlying software keeps getting more accurate. Face recognition technology that works well is not automatically face recognition technology that is being used fairly.

Cost is another disadvantage that rarely makes the headlines but matters enormously to any organization deploying this software. It is expensive to build and maintain a facial recognition pipeline that includes proper access control, encrypted storage for biometric data, staff training, and regular accuracy audits across different lighting and demographic conditions. Cutting corners on any of those pieces to save money is precisely how a casino ends up with a security system that flags the wrong person with total confidence. Organizations that treat facial recognition as a cheap plug-and-play security fix, rather than software that requires ongoing investment, are the ones most likely to end up in a Killinger-style lawsuit.

Identifying people through facial recognition is fundamentally different from identifying them through a document check, because a face can't be revoked or reissued the way a driver's license can. When facial recognition software misfires while identifying people at a casino, retail store, or public event, the person misidentified has no way to simply get a new face and start over, the error follows them. That permanence is part of why courts and legislators are increasingly skeptical of treating a facial recognition match as sufficient grounds for detention on its own. Any investigator or security professional identifying people with the help of this software should build in a manual verification step precisely because the consequences of getting it wrong are so hard to undo.

Surveillance is the broader context that all of these individual disadvantages sit inside. A single facial recognition camera at a casino door is a security tool; a citywide network of recognition cameras feeding a shared database is a surveillance infrastructure, and the line between the two is thinner than most people realize. As more retailers, transit systems, and public agencies adopt this software, the cumulative surveillance footprint grows even if no single deployment was designed with mass surveillance in mind. Investigators and security teams should be honest with themselves about which side of that line their own use of facial recognition technology falls on, because regulators increasingly are.

Face recognition, at its core, is a probability engine dressed up to look like a certainty machine, and that framing gap is the root of nearly every disadvantage discussed here. A face recognition score of 100% describes how confident the software is in its own calculation, not how likely it is that the calculation is correct in this specific real-world instance. Until that distinction is built into how investigators, security teams, and courts treat face recognition output, cases like Killinger's will keep happening. The fix isn't more confident software, it's less confident humans, willing to treat every match as a lead rather than a verdict.