Why Michigan Narcotics-Detection Dogs Trained on Marijuana Need a Hard Look After People v Armstrong
Several years into Michigan's recreational-marijuana era, I am still asked some version of the same question on a regular basis: if marijuana is legal, why did a drug dog give the police a green light to tear my car apart? It is a fair question, and after the Michigan Supreme Court's decision in People v Armstrong, ___ Mich ___ (2025), it is also a question with real legal teeth. In my practice, I now look at narcotics-detection-dog cases in a different light. The same animal that was trained and certified five or ten years ago to alert to marijuana, cocaine, methamphetamine, heroin, and ecstasy is, in 2026, almost certainly still trained to alert to a substance that adults in Michigan may lawfully possess. That single fact has consequences for probable cause, for reliability, and for whether a dog's alert means anything at all.
What Armstrong Decided
In People v Armstrong, the Michigan Supreme Court held that the rule announced in People v Kazmierczak, 461 Mich 411 (2000) — that the smell of marijuana alone could establish probable cause to search a motor vehicle under the automobile exception — is no longer good law in light of the Michigan Regulation and Taxation of Marihuana Act, MCL 333.27951 et seq. The Court explained that when Kazmierczak was decided in 2000, possession, use, and transportation of marijuana were criminal in Michigan without exception, so the smell of marijuana indicated a substantial chance of criminal activity. After voters passed the MRTMA in 2018, marijuana possession and use were generally decriminalized for adults aged twenty-one or older.
The new rule is more nuanced than a simple reversal. Armstrong recognized that certain marijuana-related conduct remains unlawful — operating a motor vehicle under the influence of marijuana under MCL 333.27954(1)(a) and MCL 257.625(1), consuming marijuana in a public place under MCL 333.27954(1)(e), and smoking marijuana within the passenger area of a vehicle on a public way under MCL 333.27954(1)(g). The smell of marijuana therefore remains a factor that may play a role in the probable-cause analysis, but it is not, on its own, sufficient. Other inculpatory facts — for example, observations suggesting intoxication or the presence of smoke — are needed to support a search for contraband. The Court tied this conclusion to the Fourth Amendment principle articulated in Ybarra v Illinois, 444 US 85 (1979), that probable cause must be particularized to the person, place, or thing to be searched.
I explain this to clients in plain terms. The smell of marijuana is now a clue, not a verdict. It can combine with other facts to support probable cause, but standing alone it does not support a warrantless search of a vehicle under the automobile exception.
Why Armstrong Reaches the Drug-Detection Dog
Police service dogs deployed for narcotics detection in Michigan have for decades been trained to alert to a panel of target odors that ordinarily includes marijuana along with cocaine, methamphetamine, heroin, and other controlled substances. When I evaluate a case, the first thing I want to know about the dog is the list of trained odors. In nearly every Michigan drug-dog case I have reviewed, marijuana is on that list, and the dog's signal — typically a passive sit or a more active scratch — does not distinguish among the trained odors. The dog tells the handler that one of the trained substances is present. It does not tell the handler which one.
This is the practical problem that Armstrong brings to a head. If a Michigan narcotics-detection dog is trained to alert to marijuana, the dog's generalized alert is now consistent with a substance that the occupant of the vehicle is lawfully permitted to possess. After Armstrong, that ambiguity is constitutionally significant, because the presence of marijuana — without more — does not give an officer probable cause to search.
Armstrong itself does not address canine training, and I do not read it as compelling, by its own force, the wholesale recertification of Michigan drug dogs. That is an inference I draw from the decision, not a holding. But the inference is, in my view, a sound one: an evidence-gathering tool whose output cannot reliably distinguish lawful from unlawful conduct should not, without further validation, be treated as the foundation for probable cause. The cleanest way to restore that foundation is to extinguish the marijuana imprint and to validate new dogs on the panel of substances that remain unlawful. Dog certified prior to 2025 should not be used in Michigan after the Armstrong decision.
What the Empirical Research Shows
While Armstrong is a Michigan decision, the science of detection-dog reliability is international, and a recent article in the Alberta Law Review collects much of it: Sara Gordon, "Call Off the Dogs: Rethinking Sniffer Dog Searches in Canadian Criminal Law," Alberta Law Review (2026). The article is written about Canadian Charter law, and its legal conclusions are therefore persuasive and comparative rather than binding in Michigan. Its survey of the empirical literature, however, is directly relevant to how I evaluate any drug-dog case under Michigan law.
The article catalogs a striking body of evidence that detection-dog alerts in the field are far less accurate than the public — and many courts — have assumed. A two-year New South Wales review of more than 10,000 alerts by seventeen drug-detection dogs found that searches following an alert located illegal drugs only about twenty-six percent of the time, with individual dogs ranging from a high of fifty-six percent down to a low of seven percent. Later New South Wales data showed false-alert rates above sixty-seven percent across multiple years. North American data is no more flattering. A 2011 Chicago Tribune investigation cited in the article reported that drug-dog alerts during traffic stops led to the discovery of drugs or paraphernalia only about forty-four percent of the time. A Louisville analysis found that forty-five percent of dog-alert searches turned up no narcotics. The Seventh Circuit memorably described one Illinois dog's field-accuracy rate as not much better than a coin flip in United States v Bentley, 795 F3d 630 (CA 7, 2015).
The article identifies several mechanisms behind these false alerts that I think are particularly important for Michigan practitioners to understand. First, dogs frequently alert to residual odors — minute traces of a target substance left behind on clothing, currency, or upholstery. The article notes published evidence that a large majority of circulating Canadian banknotes contain trace cocaine, an observation echoed in Michigan-relevant cases through long-recognized findings about cross-contamination of paper currency. Second, dogs can alert to chemicals associated with target substances rather than the substances themselves. Methyl benzoate, for example, is the dominant odor signature of street cocaine and is also a common ingredient in shampoo, soap, and perfume; piperonal is the dominant signature of MDMA and is also used in cherry and vanilla flavoring and in some insect repellents. Field studies have documented dogs alerting on piperonal samples that contained no MDMA at all.
Third, and most overlooked, dog performance is not stable. The article reports findings that approximately thirty-five percent of detection dogs temporarily lose their sense of smell each year because of illness, dental problems, or other physical issues, and that performance varies with time of day, temperature, humidity, and search environment. A 2014 Polish study cited in the article found accuracy rates of about 83.2 percent for indoor searches, 63.5 percent for the outside of vehicles, and 57.9 percent for the inside of vehicles — the very environment in which Michigan drug dogs are most often deployed.
The Clever Hans Effect and Handler Cueing
The fourth mechanism may be the most important, and it is the one I focus on most heavily when I cross-examine a handler. Detection dogs are highly attuned to subtle, often unconscious cues from their human partners — changes in posture, leash tension, breathing, voice, pace, and gaze. The phenomenon is known by reference to Clever Hans, a horse who appeared to perform arithmetic but was in fact responding to involuntary cues from his owner. The Gordon article details a now-well-known 2011 study by Lit, Schweitzer, and Oberbauer in which eighteen handler-dog teams searched rooms that, unknown to the handlers, contained no target scent at all. The handlers were told that some target locations would be marked with a piece of red paper. The dogs falsely alerted in those rooms a total of 225 times — roughly eighty-five percent of the time — and were twice as likely to alert at locations marked with red paper than at locations containing only a decoy scent. The study attributed the false alerts to handler beliefs rather than to canine olfaction.
A 2020 study cited in the article reached a complementary conclusion: detection dogs searched longer when handlers believed unfound odors were present and shorter when handlers believed there were none. The methodological implication is straightforward. Single-blind testing — in which the handler does not know where the target is hidden but an evaluator on scene does — remains, according to the article, the most common form of certification testing. Double-blind testing, in which neither the handler nor the evaluator knows the target's location, is what the literature describes as the gold standard. In real-world deployment, no one knows where the target is. Only double-blind testing replicates that condition.
How I Apply This in Michigan Drug-Dog Litigation
When I evaluate a Michigan case in which a dog alert is the linchpin of the search, I now ask the same questions in every file. What were this dog's trained odors at the time of the deployment? When was the dog last certified, and by whom? Was the certification testing single-blind or double-blind? What is the dog's documented field-deployment record, including alerts that did not lead to the discovery of contraband? How does the agency record and characterize unsuccessful alerts in its reliability statistics? Has the agency continued to train the dog on marijuana after the MRTMA, and if so, why?
These are not abstract concerns. Armstrong places the burden on the prosecution, consistent with People v Reed, 393 Mich 342 (1975), to justify a warrantless search. When the only basis offered for a search is a dog alert and the dog is trained on marijuana, the prosecution must explain how that alert gave the officer probable cause to believe contraband — as opposed to a lawful quantity of marijuana — was present. Where the record does not answer that question, a motion to suppress is the appropriate vehicle. The motion does not promise a result. It frames the legal question, identifies the gaps in the prosecution's record, and, in many cases, forces an evidentiary hearing on training, certification, deployment, and reliability that has never previously been tested in court.
Conclusion
The pairing of People v Armstrong with the empirical record collected in the Gordon article suggests that Michigan has reached an inflection point in narcotics-detection-dog jurisprudence. The legal premise that supported decades of dog-based searches — that a generalized alert reliably indicated the presence of contraband — no longer fits a Michigan in which adult marijuana possession is generally lawful. The science, drawn from sources outside Michigan but methodologically serious, raises questions about handler cueing, residual odors, alerts to legal substances, and certification standards that Michigan courts have not yet had a sustained opportunity to address. Each case must be evaluated on its own facts and its own record, and no motion is guaranteed to succeed. What I can say is that, when I evaluate a case in 2026, I now look as carefully at the dog as I do at the officer — and after Armstrong, careful counsel should do the same.
