On July 1, 2012, Milton Hall, a homeless man with a history of mental illness, stole a cup of coffee from a convenience store in Saginaw, Michigan. The store’s clerk called 911. When an officer arrived, Hall produced a knife with a three-inch blade and threatened her with it. She called for backup and seven other officers soon joined her, one of them with a police dog. They formed an arc around Hall and aimed their firearms—pistols and a rifle—at him. The standoff continued for several minutes, with the officers repeatedly asking Hall to put the knife down and Hall repeatedly refusing. Finally, Hall, still wielding his knife, began to walk toward the police dog and the K9 officer. After he had taken a few steps—three, by my count, as I watch video footage from a patrol car’s dashboard camera and available on YouTube—the officers shot Hall to death in a volley of 47 bullets.

Whether a person is dangerous, and how dangerous he is, is rarely easy to determine. Combine that uncertainty with a low tolerance for risk on the part of both officers and the use-of-force laws that govern their actions, and the result is speculative police killings: cases in which people are made to forfeit their lives on the basis of little more than guesswork about what they might do.

Such deaths are intolerable. Debates over police conduct often pit the right against the left, but the conviction that no one should die on the basis of a hunch should be trans-partisan. It is also wise to feel a certain mistrust of government power exerted against the citizen; a certain fear that agents of the state may overstep their authority or use well-intentioned but destructive zeal; a certain expectation that the legal institutions to which they are accountable may fail to hold them to account; and a certain recognition that rules defining the rights of citizens and the scope of government power against them may be vague or elastic in ways that can harm the individual. That the individuals at issue here have often violated the law may make them harder to sympathize with, but for this very reason it is proper to wonder whether the insecurity of their rights is too readily tolerated. After more than a year of debate and protest and occasional riots in response to particular police killings, it would be well to take a scrutinizing look at use-of-force rules themselves.

There is a moral logic common to such rules. Two requirements must be met for a use of force to be justified: a “proportionality” requirement and a “necessity” requirement. The proportionality requirement, as Paul H. Robinson, an expert on criminal law at the University of Pennsylvania Law School, summed it up for me, concerns whether a person posed a danger to which the force used against him was a commensurate response. For deadly force, the usual standard is that someone must present a threat of death or serious bodily injury. The necessity requirement concerns whether, at the time when deadly force was used, it was truly needed to prevent those threats from being realized. If someone makes a death threat, for example, he has satisfied the proportionality requirement, but that doesn’t mean the police may shoot him dead at once. Deadly force becomes permissible only when and if it is needed to keep him from killing.

As an abstraction, this makes sense. In practice, it can be vexingly indeterminate. A use of deadly force is, after all, preventative and the public’s judgment of it retrospective. What would definitely show it to have been justified is an affirmative answer to the question, “If deadly force had not been used against this person, would he have gone on to inflict death or serious injury?” And the answer to that question is unknowable. Sometimes the public may feel confident enough saying yes. But suppose it says, “Maybe. Who knows?” Is that reason enough to have ended someone’s life?

​* * *

When Michael T. Slager, a police officer in North Charleston, South Carolina, shot Walter Scott dead, there was universal agreement that the act was an outrage. Scott had been unarmed and running away from Slager when Slager fired eight rounds at Scott’s back. No one could have argued that using deadly force had been either proportional or necessary. But Scott’s case was perhaps uncommonly clear-cut. The killing of Milton Hall illustrates the ambiguities that can arise in applying a use-of-force standard, and the ways in which those ambiguities can lead to horrifying results.

By brandishing a knife, refusing to put it down, and approaching the officers, Hall satisfied the proportionality requirement. His words during the standoff—e.g., “Some motherfucker is gonna die today, and it’s gonna be me or it’s gonna be you”—further established that he posed a threat of death or serious injury. And yet the officers had two means less deadly than firearms—the police dog and their Tasers—available to stop him, and they did not try to use them. So was lethal force truly necessary?

I asked Chris Gebhardt, a former lieutenant with the Metropolitan Police Department in Washington, D.C., and later a SWAT-team leader in Utah, what he thought. “The officers had a K9 there,” Gebhardt answered. “Release the dog and let it do its job.” By choosing instead to open fire, he explained, “they have basically said that a dog’s life is more valuable than a human’s.” Indeed, in the moments immediately before his death, Hall seems mainly to have been confronting the dog. Michael Thomas, Saginaw County prosecutor at the time, conceded as much when he announced at a press conference that he would not file charges against the officers. (Hall’s mother later sued the city and the officers for wrongful death and received a settlement of $725,000.) “This dog handler and this dog in particular seem to be what [Hall] was directing his attention to,” Thomas said.

Thomas emphasized that, according to two witnesses, “the police dog and the police-dog handler” were two to three feet away from Hall at the closest point. But they could not have been equidistant from him, since the dog stood between the handler and Hall. As the video makes apparent, the handler began to walk the dog backward as Hall approached, maintaining his distance from Hall; and when the shots were fired, Hall was clearly more than two or three feet away from the dog. Thomas’s failure to acknowledge any of these facts supports Gebhardt’s observation that the dog’s life was considered more valuable than Hall’s. (Thomas did not respond to an email or a voice message requesting comment.)

As for Tasers, the officers had them and even considered using them: Thomas noted at his press conference that the camera and audio recorder on one Taser had been activated during the standoff, and he said that one of the officers had not fired any rounds because he was moving his hand from his Taser to his firearm when the other officers began to shoot.

Still from patrol-car dashcam footage moments before officers fire at Milton Hall (Michigan ACLU)

It would be imprudent to lay down a rule that an officer being confronted by a knife-wielding individual must always use a Taser or release a dog before shooting a firearm. Tasers in particular don’t always stop their targets, as Thomas emphasized, and someone sprinting at you with a knife can cover a lot of ground—21 feet, according to an influential police study—in the time it takes to draw a pistol. But the officers had their pistols drawn already. Hall was not sprinting, but walking normally. And the officers need not have waited until Hall approached them to try to subdue him with less deadly means. Here is Gebhardt again: “Critical thinking . . . would have caused a two-man team to approach him. One with a Taser and another with lethal cover. Get close enough to use the Taser. If he approaches with the knife, then shoot.”

Thomas’s reasoning focuses on two details almost to the exclusion of all others: Hall’s possession of a knife and the officers’ lack of certainty that non-lethal means would have stopped Hall. If that is all you look at, and if you have little tolerance for risk, then it’s easy to think that the officers needed to kill Hall. But the more you look at the overall context, and the more you reflect on how the officers might have approached the situation instead, the less obvious that necessity looks.

​* * *

This illustrates something important about judging uses of deadly force: It requires attending to the totality of circumstances. The Supreme Court has emphasized this point. Uses of force are considered seizures of someone’s person and so are subject to the Fourth Amendment’s prohibition of unreasonable searches and seizures. And the test of whether a use of force is reasonable, the Court said in its 1989 Graham v. Connor decision, “is not capable of precise definition or mechanical application.” Instead, “its proper application requires careful attention to the facts and circumstances of each particular case.” The more panoramic the view, the less risk that isolated details will skew judgments.

Nevertheless, the Court had already set up a mechanical test prior to deciding Graham v. Connor. In 1985, Tennessee v. Garner posed the questions whether and when a police officer may use lethal force to seize a fleeing suspect. Under common law, officers had been allowed deadly force against any fleeing felon if using it was necessary to prevent his escape. Garner involved a case in which a police officer acting on that standard—which had been enshrined in Tennessee’s statues—fatally shot a fleeing, unarmed burglar in the back of the head as the burglar climbed a fence. The Court’s majority noted that the common-law standard had developed in times when police officers did not carry firearms and deadly force implied hand-to-hand struggle. It had also developed when almost all felonies were capital offenses. In the modern context, the Court argued, “it is not better that all felony suspects die than that they escape. Where the suspect poses no immediate threat to the officer and no threat to others, the harm resulting from failing to apprehend him does not justify the use of deadly force to do so.”

Notice where the Court placed the word “immediate.” It appears before “threat to the officer” but not before “threat to others.” That is not a result of careless composition. It captures the idea that some suspects pose an ongoing danger even if, in the moment of their escape, they are not threatening anyone in some specific way. The opinion continues:

When the officer has probable cause to believe that the suspect poses a threat of serious physical harm, either to the officer or to others, it is not constitutionally unreasonable to prevent escape by using deadly force. Thus, if the suspect threatens the officer with a weapon or there is probable cause to believe that he has committed a crime involving the infliction or threatened infliction of serious physical harm, deadly force may be used if necessary to prevent escape, and if, where feasible, some warning has been given.

In such cases, the police do not need any specific reason to think that the suspect will threaten death or injury in the future. What he has already done is taken as reason enough to prefer his death to his escape. The problem with this is that there is no reliable way to predict someone’s future actions unless he has manifested an intention to do something. A fleeing suspect has manifested an intention to flee. The probability that he in particular—as opposed to a hypothetical average individual who acts according to a statistical mean—will commit a violent crime is unknowable. The Court’s rule, in short, is entirely speculative. And that should raise doubts about whether it is a reasonable way to define proportionality.

​* * *

On September 10, 2014, Darrien Hunt, who had a history of emotional problems and had shown signs of mental illness, was seen walking with a samurai sword down a busy street in Saratoga Springs, Utah. According to his family, he carried the sword as part of a costume, and under Utah law he had a right to carry any sharp-edged weapon he wished.

A concerned observer called 911, and two police officers approached Hunt. The conversation started off friendly, but after one of the officers repeatedly asked Hunt to put down his sword, Hunt either took a swipe at the officers or held the sword over his head and jumped at them. (The officers gave investigators conflicting accounts of what Hunt did with his sword and which officer he was moving toward when he did it.) The officers responded by shooting at Hunt, who ran away carrying the sword. The officers ran after him, and one of them shot Hunt dead from behind.

Hunt’s family is suing the officers and the city of Saratoga Springs. Many facts about the incident are contested—in particular, whether Hunt had already dropped his sword, and whether he was tripping or had fallen down, when the fatal shots were fired. Witness testimony is inconsistent, and lawyers for both sides declined to answer questions about the ambiguities. But, assuming for the sake of argument that something like the officers’ accounts is accurate, what difference should it have made that Hunt was still on his feet and carrying a sword when he died?

Utah’s relevant statutes apply with exactness the rules in Tennessee v. Garner, but, interestingly, not even the officers’ own attorney is defending them on the grounds that Hunt had threatened them with a weapon and so satisfied the Garner standard. In a press statement, she said that, “had Hunt dropped the sword before fleeing, the officers would have continued chasing him on foot” rather than shooting him. Thus she tacitly conceded that it matters whether Hunt was still armed when he died.

Others likewise stressed the imminence of the threat that Hunt allegedly posed. Jeff Buhman, the Utah County prosecutor who declined to file charges against the officers, said: “I don’t find it reasonable to require that [officers] permit a person who is armed and has most immediately attempted to wound or kill police officers to escape into a presently populated retail area.” (Hunt was fleeing in the direction of a Walmart.) And Matthew Schauerhamer, the officer who killed Hunt, explained to investigators that “there was no way around it. I couldn’t keep letting him run around with a frickin’ sword.” Hunt might “hack the first person he saw.”

Some police departments have made imminence central to their use-of-deadly-force policies, thereby setting a stricter standard than Garner does. The New York Police Department’s patrol guide, for example, says that “police officers shall not discharge their firearms to subdue a fleeing felon who presents no threat of imminent death or serious physical injury to themselves or another person present.” Police policy in Philadelphia is similar.

But what does “imminent” mean? Something very expansive, it would seem, for those who think the shooting of Hunt justified. Hunt had not threatened anyone besides the officers, as the officers knew, and he had given no indication of what he would do after he fled. The officers’ rationale amounted to: “This guy could do anything.” Implicitly, such an understanding of imminence tightens the Garner standard only by requiring that the fleeing suspect be armed, and perhaps also that his initial crime immediately precede his flight.

Gebhardt, the former SWAT-team leader, has reviewed evidence from the shooting at the Hunt family’s request. (He does not do such work commercially and is not paid as an expert witness.) He interprets the idea of imminence very differently. “Once [Hunt] starts running, I see no justification for shooting him,” he says. “Darrien lashed out at authority,” but “he [had been] seen walking in the area with the sword and he didn’t attack anyone. It was the officer who thought Darrien was going to do others harm.” The shooting would have been justified, Gebhardt implies, only if Hunt had shown an intention to harm the public.

This points to a basic and unavoidable dilemma: whether to require specific evidence of someone’s violent intentions before allowing the use of lethal force, or, alternatively, to treat past violent conduct as reason enough. Call these the restrictive and the permissive approaches. The restrictive approach makes it all but certain that among those the police let escape will be some who go on to inflict harm in ways that could not have been reliably predicted. The permissive approach makes it all but certain that among those the police kill will be some who would not have gone on to inflict any harm whatsoever.

I shared my misgivings about the wide use of permissive standards with Robinson, the law professor. “I think you underestimate the damaging effect of the criminal-justice system being unable to effectively arrest suspects,” he replied. “Yes, there is a societal cost when lethal force is used against arrestees who flee, but there is also a societal cost to the system’s reduced ability to arrest persons for past violent crimes.”

This kind of utilitarian analysis, however appropriate it may be in other areas of the law, seems at odds with the very idea of a constitutional right. Such rights function to protect individuals from the state, even, or especially, when the state might find it useful to abridge them. A standard allowing police officers to kill fleeing felons under the conditions set by Garner is useful, since some of those felons—who knows which?—will go on to commit violent crimes. But no one can say with confidence that any particular fleeing felon will be the one to strike. And if it is therefore disproportionate to kill him, the current Fourth Amendment standard is not reasonable.

On the other hand, perhaps a lower standard of proportionality is appropriate for someone who has “put himself on the wrong side of the law,” as Paul G. Cassell, a former federal judge and a professor at the University of Utah’s law school, put it to me. By definition, a fleeing felon against whom Garner would authorize deadly force has—so an arresting officer has probable cause to believe—committed a violent crime, or else threatened an officer with a weapon. He has then run away when the police tried to stop him. Quite apart from any utilitarian analysis of the social consequences of a Garner-like standard, these factors could be taken to justify the use of deadly force against him.

But the deaths of Milton Hall and Darrien Hunt, among others, show that Garner set the bar too low. At a minimum, police officers should be allowed to use lethal force only when it is reasonable to believe that a suspect is armed. This would temper speculation about future conduct with at least a loose notion of imminence. Second, police officers should not be allowed to shoot a fleeing felon unless he has threatened a member of the general public—and not merely a police officer—with death or serious injury. The police should of course try to arrest those who threaten them with weapons, and in doing so they must defend themselves as necessary. But to assume that such a person will also threaten members of the general public requires a large speculative leap—not only from past conduct to future conduct but also from one kind of circumstance to others that are very different.

​* * *

There is one class of police shootings in which the proportionality requirement seems to have been all but abandoned: car chases. The Supreme Court has given officers wide latitude to kill drivers simply on the fear that they will cause traffic accidents.

It’s indisputable that someone driving recklessly to escape the police poses a serious public-safety threat. A recent analysis conducted by USA Today found that, since 1979, 2,456 bystanders have been killed by police chases, along with 2,610 people who had not violated the law but were passengers in cars under pursuit. Because these findings were based on police reports, which sometimes omit mention of a chase, the true numbers are probably higher.

Yet it is hard to believe that the best response is for police to open fire on fleeing drivers. As a matter of principle, a reasonable standard of proportionality should match intent with intent: Officers should not try to kill people unless they are trying, or at least have tried, to kill or injure others. Dangerous though he be, a reckless driver presents a categorically less severe kind of threat than someone who has shown that he aims to inflict harm. And as a practical matter, if a driver is disabled, “the vehicle will most likely continue under its own power or momentum for some distance[,] thus creating another hazard,” as the International Association of Chiefs of Police has noted.

The Supreme Court’s sole reasonable car-chase decision came in the 2007 case of Scott v. Harris. There, the Court ruled 8–1 that an officer who had forced a fleeing driver off the road, inadvertently causing a crash that paralyzed him, had acted consistently with the Fourth Amendment. If safer means of stopping the car were unavailable—a claim that Justice John Paul Stevens disputed in his dissent—this conclusion is sound. But the Court has also found it constitutional to shoot (and kill) a driver whose flight had been momentarily interrupted, lest he start driving recklessly again. And it has even granted immunity from civil liability to an officer who shot a fleeing suspect just as his car began to move—that is, before the officer had had any opportunity to observe whether the suspect’s driving might pose a threat.

In a ruling on that case that the Supreme Court overturned, the Ninth Circuit Court of Appeals had observed that police officers are free to discontinue car chases that place the public at risk. As Judge Stephen Reinhardt put it in a concurring opinion, “the chase itself cannot create the danger that justifies shooting [the] suspect.” The Supreme Court later countered, in Scott v. Harris, that discontinuing a pursuit does not guarantee that a suspect will stop driving dangerously. But allowing deadly force to end chases in effectively all circumstances takes this logic to the opposite extreme.*

Ending a chase may not guarantee that a suspect will drive less dangerously. But most suspects are likely to slow down shortly after officers cease their pursuit, as studies based on interviews with apprehended suspects and aerial observations of discontinued police chases have found. And, provided that later attempts will be made to identify and arrest a suspect who escapes, it is simply not true, pace the Court’s majority in the same opinion, that ending dangerous car chases creates “an invitation to impunity-earned-by-recklessness.” This reasoning falsely assumes that a policy of shooting reckless drivers is the only possible deterrent, when in fact criminal penalties for fleeing a police officer act as deterrents as well.

Many police departments have implemented policies more restrictive, and more reasonable, than the Court’s breathtakingly permissive standard. In an amicus brief filed in Scott v. Harris, the Georgia Association of Chiefs of Police said that “when the immediate danger to the public created by the pursuit is greater than the immediate or potential danger to the public should the suspect remain at large, then the pursuit should be discontinued or terminated ... Pursuits should usually be discontinued when the violator’s identity has been established to the point that later apprehension can be accomplished without danger to the public.” And the NYPD’s patrol guide says that “police officers shall not discharge their firearms at or from a moving vehicle unless deadly physical force is being used against the police officer or another person present, by means other than a moving vehicle.” The last clause may be taken as a rebuke to officers who have shot drivers whom they claimed were trying to run them over—as if doing so would magically stop the car.

Of course, like all policies on the use of deadly force, this one creates ambiguities. If, for example, officers hear a single noise that might be a car backfiring or might be a gunshot, may they discharge a hail of bullets at the car’s occupants at the first possible instant? Such were the circumstances in which Timothy Russell and Malissa Williams, both unarmed, died needless deaths in Cleveland.

* * *

It is astounding that some police departments have voluntarily set stricter deadly-force standards than the courts have required them to employ. Often it is law-enforcement agencies that push the bounds of the permissible and courts that rein them in; here the relationship is reversed. The ways in which some departments have tightened their rules may suggest ways in which the constitutional interpretation itself will eventually improve. (The Constitution is susceptible of evolving interpretation on this point, since the Fourth Amendment says no more than that seizures may not be unreasonable.)

But there will be limits to what legal reforms can accomplish. In criminal law, proof beyond reasonable doubt establishes a suitably high bar for prosecutors to clear. And in civil law, the doctrine of qualified immunity shields officers from liability for violating people’s constitutional or statutory rights—and usually entitles them to summary judgment in their favor—if it can be shown that the legal standard was not clearly defined. In the murky, context-dependent area of Fourth Amendment case law, it often isn’t. This makes it very hard to impose on police officers any form of accountability between criminal guilt and full exoneration.

Legislatures could partially address that problem by establishing that officers will not be granted qualified immunity in wrongful-death suits when their actions violate departmental training or policy. And independent bodies whose members include community representatives could be empowered to fire officers even if they have not been found criminally culpable or civilly liable.

But those are retrospective remedies. Courts and investigative bodies sort through the consequences as well as they can, but what they cannot do is bring back the dead. So the best solution must include widespread police training that emphasizes deescalation and helps officers win compliance before they ever consider using deadly force.

It isn’t hard to see how such an approach might have saved the life of Milton Hall. “The dog was unnecessary,” says Mark Fancher, an attorney with the Michigan ACLU who tried unsuccessfully to get the Justice Department to file civil-rights charges concerning Hall’s death. “[The police] were not going to unleash him,” says Fancher. “All the dog served to do was get [Hall] more agitated.” Fancher also wonders why the officers formed what was effectively a firing squad around Hall in the first place. Could they not have placed themselves at a distance, kept bystanders away, and, monitoring Hall, given him a chance to calm down?

No permission granted by the people to agents of the state is of graver consequence than permission to take life. It is often exercised under the sway of adrenaline and powerful emotion. By pitting the state against the citizen, it carries with it the ineliminable possibility that abuse and caprice will not be held properly to account. Given all of this, it is difficult to justify authorizing the use of deadly force for anything but unambiguous and compelling reasons. This principle seems to be at work in the public’s reactions to recent police killings. Perhaps legislators, the courts, and the police are ready to apply it.

* This article originally implied that Judge Stephen Reinhardt's opinion disregarded the danger that  fleeing suspects might pose after the police allow them to escape. In fact, Reinhardt's opinion is compatible with acknowledging that danger. The article has been clarified to more accurately represent the Ninth Circuit's ruling. We regret the error.