SUTTON, J., delivered the opinion of the court in which NORRIS, J., joined, and DONALD, J., joined in part. DONALD, J. (pp 236-44), delivered a separate opinion concurring in part and dissenting in part.
OPINION
SUTTON, Circuit Judge.After Robert Mitchell resisted arrest by the police, an officer shot two taser darts into his chest. Robert experienced cardiac arrest and died shortly after. His mother filed this lawsuit in response. It proceeded in two phases — first against the individual police officers and the City of Warren, then against the weapon’s manufacturer, Taser International, Inc. The parties settled the claims against the police officers and the City. The other claims are the subject of this appeal. Below, the district court granted summary judgment to Ta-ser, rejecting the plaintiffs failure-to-warn and negligence claims as a matter of law. We affirm.
I.
On the morning of April 10, 2009, Jesse Lapham, an officer with the City of Warren Police Department, responded to a call for backup at an abandoned house near Eight Mile Road. As explained by the radio dispatcher, one of Lapham’s fellow Warren police officers had stopped a ear with expired tags when one of the passengers — 16-year-old Robert Mitchell — ran from the car, broke into an abandoned house, and hid upstairs. Lapham arrived at the house to the sound of the pursuing officer coaxing Robert to come down without a fight.
Robert complied, and an officer began to arrest him once he arrived downstairs. At *226that point, for reasons that the record does not disclose, Robert tried to evade the officer’s grasp, and a struggle ensued. Lapham de-holstered his taser and shot Robert with it. One metal dart hit just inches above Robert’s heart, the other just inches below. Mitchell fell to the ground. When paramedics arrived on the scene, they detected “v-fib”: ventricular fibrillation, “a highly disorganized heart rhythm ... at rates of 400-600 [beats] per minute” that results in cardiac arrest if uncorrected. R. 236-46 at 18; see R. 236-22 at 1-2. The medical team tried to resuscitate him but could not.
The Warren Police Department purchased this X26 taser in August 2006. In mid-September 2006, several Warren officers attended a taser-training school run by an instructor hired by Taser. The Warren officers learned to aim the X26 at “center mass” because that presents “the best spot to stop a person” from advancing and a head shot poses undue risks to the individual. R. 236-41 at 3; see R. 236-16 at 14. They were told that officers should, “if practicable, deploy [the] X26 at [a] suspect’s back,” as opposed to the front, but that, even when the taser’s darts land on the chest, the weapon is safe. R. 236-18 at 179. Lapham learned to use a taser from a Warren training officer. The same basic training materials were in place when he joined the Warren Police Department in August 2007.
After Robert died in April 2009, his mother Cora filed this lawsuit. She first proceeded against the City of Warren, Officer Lapham, and several other Warren-police officers. She sought recovery for Robert’s death on several grounds: use of excessive force, racial discrimination, inadequate officer training, interference with familial relations, and wrongful death. The officers moved for summary judgment, which the district court granted in part and denied in part. At that point, Mitchell settled all outstanding claims against the officers and the City.
Meanwhile, she amended her complaint to add claims against Taser. Through the amendment, she sought additional liability on two grounds: that the company negligently failed to warn end-users about the risk of cardiac arrest from an X26 discharge to the chest, and that the company’s training program negligently assured end-users that the device was safe when aimed at the chest. The district court granted summary judgment for Taser. It ruled that (1) the company had no duty to warn the Warren Police Department about any cardiac risks at the time of sale in August 2006, (2) Michigan law precluded any post-sale duty to warn, (3) Taser had not assumed a duty to warn by virtue of its training regimen, and (4) Mitchell could not prove that Lapham would have ever seen a warning even if Taser had issued one.
II.
Mitchell first challenges the district court’s duty-to-warn ruling. In doing so, she does not claim that the safety costs of using a taser exceed its benefits — that the company was negligent in making the product for law-enforcement purposes. Nor does she claim that Taser was negligent in recommending that officers aim the darts at the torso of a criminal suspect. What she claims is that the company should have warned police departments about the health risks of shooting the darts into the upper part of the front torso — the chest. In addressing this claim, we look to a state statute, not the common law. The Michigan legislature has codified a product manufacturer’s duty to warn end-users about dangers associated with a product’s use. The relevant statute provides:
*227In a product liability action brought against a manufacturer or seller for harm allegedly caused by a failure to provide adequate warnings or instructions, a manufacturer or seller is not hable unless the plaintiff proves that the manufacturer knew or should have known about the risk of harm based on the scientific, technical, or medical information reasonably available at the time the specific unit of the product left the control of the manufacturer.
Mich. Comp. Laws § 600.2948(3). Any liability for failure to warn in this case thus would require a showing that Taser “knew or should have known” about the risk of cardiac arrest from an X26 chest shot based on the information available “at the time” of sale — here August 17, 2006. A reasonable jury could not make such a finding for several reasons.
Field use of tasers. Tasers first became available to officers in 1994. The trademarked acronym has a longer heritage. It stands for “Tom A. Swift electrical rifle,” named after the young-adult novel that later inspired the weapon. See Victor Appleton, Tom Swift and His Electric Rifle (1911). When discharged, a taser fires two darts, connected to the main unit by wire, that puncture the victim’s skin and send a 0.0021-amp electrical current into the body for five seconds. (A wall outlet, by comparison, delivers 16 amps; a single Christmas-tree light bulb delivers one.) The stunning effect of the five-second discharge permits police to reduce the risks of injury to individuals and officers that arise in law-enforcement encounters from time to time. Tasers permit officers to avoid using physical force in face-to-face contact to control suspects who resist arrest and permit officers to avoid using guns (and bullets) to disable or stop a dangerous and resistant suspect. In the words of one officer: “The whole point is for police to have more options to end a confrontation without further injury to the officer or the suspect.” R. 236-24 at 3. By and large, tasers have done just that. As police forces integrated tasers into their non-lethal-force policies, they saw reductions in officer injuries, suspect injuries, uses of lethal force, and complaints of excessive force. At the same time, tasers caused fewer injuries than hand-to-hand restraint, chemical sprays, or batons.
Through August 2006, the use of tasers in the field confirmed that they lessened many of the occupational hazards associated with police work and many of the risks associated with suspects who refuse to follow police orders. By the end of 2005, over 8,700 law-enforcement agencies had purchased tasers. See Taser Int’l, Inc., Annual Report 6 (Form 10-K) (Mar. 16, 2006). And by then field discharges of them had reached the hundreds of thousands. See Mark W. Kroll & Jeffrey D. Ho, TASER Conducted Electrical Weapons: Physiology, Pathology, and Law 283 (2009) (estimating 606,395 field uses of ta-sers between 2001 and 2008); Kroll et ah, TASER Safety, Canadian Med. Ass’n J. (Sep. 23, 2008), http://www.ncbi.nlm.nih. gov/pmc/articles/PMC2535727/ (estimating that tasers “ha[d] been applied to over 1.3 million people” by 2008); see also Statistics and Facts, Taser Int’l, Inc. (Mar. 25, 2015), https://www.taser.com/press/stats (estimating 2.82 million applications through July 2015).
Noncompliant suspects have not been the only people exposed to tasers. Taser’s training program allots time for officers to experience a taser discharge if they so choose. Taser’s volunteer-exposure guidelines suggest, as the company advises for all discharges, that “[p]robe hits to the back are safer” and to “avoid the face, throat, [and] groin.” R. 236-18 at 81. In the heat of law enforcement encounters, *228officers of course do not always have that choice when trying to stop or control a suspect.
Through it all, just two potential cardiac-arrest problems arising from taser use — both mentioned by Mitchell’s expert witness, Dr. Douglas Zipes — had been made public before Taser shipped the Warren Police Department’s X26s. Neither one showed that the company should have warned officers not to shoot a taser in the top half of the front torso — the chest. In one, an officer stunned — using the drive stun, as opposed to the dart, feature of a taser — an epileptic man suffering from “paranoid delusions” and “frothing at the mouth” five consecutive times in the chest while the man was strapped in a chair, held down by nine officers, and placed in a choke hold. See Notice of Removal with Complaint, Ex. 1 at 9, 28, 32-34, Williams v. TASER Int'l, Inc., 1:06-cv-00051, 2006 WL 6510516 (N.D.Ga. Jan. 9, 2006); see also Consent Motion, Williams, 1:06-cv-00051 (N.D. Ga. Apr. 21, 2009) (settling claims against Taser). The incident does not show that ventricular fibrillation occurred, as the cause of the victim’s “car-diorespiratory arrest” was “undetermined.” Global Appendix, Ex. 82 at 13-14, Williams, 1:06-cv-00051 (N.D.Ga. July 25, 2008). The other incident is more opaque. What we know about it comes from a two-paragraph letter to the editor of the New England Journal of Medicine in September 2005, saying that an unnamed adolescent collapsed after being tased and experienced ventricular fibrillation “within two minutes.” The letter does not say whether the taser darts hit the suspect’s chest or for that matter the front or the back of the suspect. The record does not contain any autopsy or any other medical data about the incident. Based on the field use of tasers through August 2006, nothing showed that Taser was negligent in the warnings issued with its product.
The medical literature. The relevant medical literature before August 2006 supported the safety of tasers as a law-enforcement option and more to the point did not establish a risk of cardiac arrest from instructing officers to aim the darts at the torsos of non-compliant suspects. The Human Effects Center for Excellence published a comprehensive review of taser safety in June 2005. It surveyed the results of five controlled animal studies— conducted by M. Nerheim, R.A. Stratbucker, and W. McDaniel, Dr. Stratbucker by himself, and O.Z. Roy and A.S. Podgor-ski — all of which suggested that the X26’s standard charge was too weak to cause cardiac arrest. One study indicated that only a current 16 times stronger than the X26’s could induce ventricular fibrillation. Another found that ventricular fibrillation did not result even after stimulating the surface of the heart directly with X26-level output. The Center concluded that, “for large children and adults, even those who might be sensitive responders, the risk of inducing [ventricular fibrillation] is very small, since a large margin of safety exists.” R. 197-87 at 12.
A January 2005 peer-reviewed article in the Journal of Pacing and Clinical Elec-trophysiology supported these findings. It found that cardiac safety depended on the target’s weight and that animal studies underestimated the safety margin for people. It added that the safety margin for pigs, the typical animal used in taser studies, was large: Ventricular fibrillation occurred in a 66-pound pig only with a taser stun 15 times greater than the standard charge. R. 197-39 at 4-5.
Two August 2006 studies — formally released week before the City of Warren received its delivery of X26s but informally conveyed to Taser several months earlier — qualified these findings but did not *229alter Taser’s duty to warn. The Journal of the American College of Cardiology published both studies, with Dr. Patrick Tchou and his mentee Dr. Dhanunjaya Lakkireddy leading one and Dr. Kumaras-wamy Nanthakumar leading the other. Taser provided a research grant to Tchou and Lakkireddy, while Nanthakumar received independent funding. Both studies, like many of the earlier studies, used sedated pigs as test subjects. Both implanted an electrode inside the pigs’ hearts for close monitoring of cardiac activity. Both placed the X26’s darts on the pigs’ chests — one directly above the heart and one directly below. And both found that, when shot at the chest, the X26 could “capture” a pig’s heartbeat. R. 236-10 at 3; R. 236-12 at 6. Cardiac capture occurs when “[t]he heart’s natural [ ] rhythm ... [is] taken over by an external electrical source.” R. 236-46 at 18. That speeds up the heart rate, and given enough time a sufficiently rapid heartbeat will devolve into ventricular fibrillation — an erratic rhythm that may lead to cardiac arrest.
The studies, however, showed little risk that the standard charge could create cardiac arrest — the source of Mitchell’s death. Tchou and Lakkireddy found that ventricular fibrillation, which if left uncorrected is likely to cause cardiac arrest, occurred only after modifying the X26 to generate a charge between 5 and 10 times more powerful than the norm. In the researchers’ words: “[The] study showed that [ventricular fibrillation] could not be induced using the standard 5-[second] [t]aser discharge applied to a pig’s body surface even at the most sensitive area tested.” R. 236-10 at 5. They thus concluded that a “discharge at the standard 5-[second] interval is unlikely to cause life-threatening arrhythmias, at least in the normal human heart.” Id. Nanthakumar managed to induce ventricular fibrillation with a standard X26 charge, but only once out of 150 tries and only after artificially increasing the pig’s resting heart rate with adrenaline to simulate a stressful encounter. He concluded that “there exists the possibility of serious [alteration of the heart’s rhythm] during [taser] discharges in structurally normal hearts during intense ... stress,” but that “[i]t is unknown whether darts positioned across the chest in humans would ... result[ ] in the same voltages” as in a pig. R. 236-12 at 6. None of the studies established to a reasonable level of certainty that an X26 poses a risk of cardiac arrest in humans. The single time one study induced ventricular fibrillation in a pig does not replace the multiple studies indicating that an X26 has no untoward cardiac effects. Neither Tchou and Lakki-reddy nor Nanthakumar in the end could do more than speculate that there is a theoretical possibility of an X26 inducing cardiac arrest in a human through cardiac capture. Mitchell admits as much: “The probability of triggering a cardiac arrest from X26-induced capture cannot be quantified.” Reply Br. 13.
As of the date of the August 2006 sale, and even after accounting for two 2006 studies released soon before the date of sale, the sum total of evidence that an X26 could induce cardiac arrest was: one instance of ventricular fibrillation occurring in one pig in one study of 150 taser discharges. That does not show that Taser should have known whether a chest shot from an X26 posed a risk of cardiac arrest in humans. The extant scientific information available to Taser indicated that the X26 at its standard charge would not induce ventricular fibrillation or cardiac arrest.
Taken together, the field use of tasers and the medical literature before August 2006 fail to create a triable issue of fact about Taser’s duty to warn. Before *230such a duty arises, the plaintiff must show that a manufacturer knew or should have known its product posed the particular risk at issue in the case. See, e.g., Rodriguez v. Stryker Corp., 680 F.3d 568, 571-73 (6th Cir.2012); Meyerhoff v. Michelin Tire Corp., 70 F.3d 1175, 1181 (10th Cir.1995); Grover v. Eli Lilly & Co., 63 Ohio St.3d 756, 591 N.E.2d 696, 699-700 (1992); Lamb v. Manitowoc Co., 570 N.W.2d 65, 69-70 (Iowa 1997). A manufacturer’s knowledge that a drug causes blurred vision, to take one example, does not prove that it should have known the drug caused retinal damage. See Basko v. Sterling Drug, Inc., 416 F.2d 417, 421 (2d Cir.1969). A manufacturer’s knowledge that exposure to a chemical damages the liver, to take another example, does not prove that it should have known exposure causes cancer. See Taylor v. Am. Chemistry Council, 576 F.3d 16, 27-28 (1st Cir.2009). And so on. See, e.g., Anguiano v. E.I. Du Pont de Nemours & Co., 44 F.3d 806, 812 (9th Cir.1995) (knowledge of danger in teflon-based hip replacements does not prove knowledge of danger in teflon-based jaw replacements); Perlmutter v. U.S. Gypsum Co., 4 F.3d 864, 869-70 (10th Cir.1993) (knowledge that some varieties of asbestos cause lung damage does not prove knowledge that all varieties did).
All that the relevant medical literature in this case showed was that a chest shot from a taser created a material risk of cardiac capture. It did not show any more than a possibility, however, that cardiac capture might cause ventricular fibrillation that could cause cardiac arrest. Robert Mitchell died from cardiac arrest. A jury may not speculate that a manufacturer should have known about one risk because a separately known risk revealed the mere possibility of the first. We have refused to allow a jury to infer, for example, that injecting one type of fluid into a joint can cause harm from the mere fact that injecting other fluids can cause harm. See Rodriguez, 680 F.3d at 571-73. If the manufacturer knows that one fluid can cause harm, it of course knows that a similar problem arising from other fluid is possible. But that does not suffice. A jury needs some evidence showing that the product in question causes the harm at issue. See id. at 573. Courts do not demand scientific certainty. See Moran v. Johns-Manville Sales Corp., 691 F.2d 811, 814 (6th Cir.1982). “[Ajrguably, there are no certainties in science.” Daubert v. Merrell Dow Pharm., Inc., 509 U.S. 579, 590, 113 S.Ct. 2786, 125 L.Ed.2d 469 (1993). But speculative risk does not equal known risk. See Carlin v. Superior Court, 13 Cal.4th 1104, 56 Cal.Rptr.2d 162, 920 P.2d 1347, 1353 (1996).
This problem of speculative leaps about product dangers also surfaces in the context of proving causation. We have refused to rely on studies establishing that a product can possibly cause an injury to prove a product probably caused the injury. See, e.g., Conde v. Velsicol Chem. Corp., 24 F.3d 809, 814 (6th Cir.1994); Turpin v. Merrell Dow Pharm., Inc., 959 F.2d 1349, 1352 (6th Cir.1992); Siegel v. Dynamic Cooking Sys., Inc., 501 Fed.Appx. 397, 405 (6th Cir.2012); see also Tamraz v. Lincoln Elec. Co., 620 F.3d 665, 670 (6th Cir.2010) (stating that speculation does not amount to “scientific knowledge” under Evidence Rule 702); Kalamazoo River Study Grp. v. Rockwell Int’l Corp., 171 F.3d 1065, 1073 (6th Cir.1999) (refusing to allow jury to “speculate” based on expert testimony to fill a “gap” in the evidence). Mere possibility, in other words, cannot establish a fact to the degree of certainty necessary to justify reliance on that fact. That is why the federal courts will not qualify an expert to testify about the safety of a product when “there is simply too great an analytical gap be*231tween the data and the opinion proffered.” Gen. Elec. Co. v. Joiner, 522 U.S. 136, 146, 118 S.Ct. 512, 139 L.Ed.2d 508 (1997).
This all makes good sense when applied to the duty to warn. A company does not have a duty to warn of all theoretically possible dangers. ' “[A]n obligation to warn against all injuries that conceivably might result from the use or misuse of a product” would make it “practically impossible” for manufacturers “to market their goods.” Glittenberg v. Doughboy Recreational Indus., 441 Mich. 379, 491 N.W.2d 208, 212 (1992). Warnings cannot serve their purpose, another Michigan court recognized, unless they “call the consumer’s attention to a danger that has a real probability of occurring and whose impact will be significant. One must warn with discrimination since the consumer is being asked to discriminate and to react accordingly,” Dunn v. Lederle Labs., 121 Mich.App. 73, 328 N.W.2d 576, 581 (1982), a point that is particularly salient when it comes to the kinds of law-enforcement encounters that require ta-sers.
The Tchou/Lakkireddy and Nan-thakumar studies both demonstrate that an X26 may in some circumstances capture a target’s heartbeat if its electric current vectors across the heart. No one disputes that. But Mitchell seeks not a warning about cardiac capture but a warning focused on “cardiac arrest and sudden death.” Appellant’s Br. 2. And cardiac capture, in the words of Mitchell’s expert Dr. Zipes, does “[n]ot necessarily” lead to ventricular fibrillation, cardiac arrest, or “lethal cardiac consequences.” R. 197-20 at 5. Knowing that an X26 causes cardiac capture raises only the possibility that it can induce cardiac arrest. To reach a jury, Mitchell needs more. She must show that Taser knew or should have known that the X26’s standard charge poses a realistic threat of cardiac arrest. But neither study goes that far. They speculate only that cardiac arrest is possible. True, Dr. Nanthakumar induced ventricular fibrillation in a pig, but he did so just once out of 150 tries and as a result concluded only that the X26 “may have cardiac risks that require further study.” R. 236-12 at 7 (emphasis added). Taser could not have been relatively confident that the X26 posed a risk of cardiac arrest in humans based on a study showing only that such a risk cannot be ruled out. At most, the studies indicate the need for more studies. A reasonable jury therefore could not find the company had a duty to warn.
Mitchell offers several responses, each unpersuasive. She first suggests that we are missing the import of the Tchou/Lakki-reddy and Nanthakumar studies. It matters not, she says, that the studies did not prove the X26 could cause ventricular fibrillation that could cause cardiac arrest. They revealed that cardiac capture could happen. And, as Dr. Zipes says, cardiac capture alone “can be life threatening, as any interference with the heart’s natural rhythm can trigger an arrhythmia.” R.. 236-46 at 19. But that is not what the two studies reveal. Capture occurred regularly in both studies, but deadly cardiac consequences did not. Nothing in the record establishes that Taser should have known a standard taser charge could speed up a human’s heart fast enough to induce cardiac arrest. And nothing in the record indicates that Taser should have known that any capture at all can be life-threatening. All Dr. Zipes could say is that cardiac capture creates heightened risks if it occurs during a “vulnerable period” of the heart’s natural rhythm, which can “cause an extraneous beat,” which can cause ventricular fibrillation, which can cause cardiac arrest. Id. at 48. But Dr. Zipes concedes that his “vulnerable period” theory *232“has not been reported so far in the animal studies.” Id.
Maybe so, Mitchell responds. But several months before the two relevant studies came out, she says, Dr. Tchou informed Taser “that despite the fact that [he] did not induce ventricular fibrillation in any of the pigs,” his research indicated “there is some possibility that [a standard X26 discharge] could induce ventricular arrhythmias in people.” R. 236-43 at 16. True enough. But the inferences Mitchell draws from this do not follow. To start, the warning she demands focuses on cardiac arrest, not arrhythmias in general. Scientists we are not, meaning we need evidence connecting the dots. No record evidence shows that the kind of arrhyth-mias an X26 could induce creates a risk of cardiac arrest. Dr. Tchou at any rate spoke only in terms of possibilities — the kind of information that at most demands more study, not the kind of information that shows Taser knew or should have known that an X26 charge could cause fatal arrhythmias in humans.
Mitchell adds that we should not discount the pig studies because they are not human studies. Yes, of course. No one is demanding a human study, least of all the individual who would be the subject of it. What is fair is to require that the study show a reasonable degree of danger after accounting for the weight and other differences between humans and pigs. Otherwise, the analogy is not helpful. No such reasonable degree of danger was shown by August 2006 when the lessons from the pig studies are extrapolated to account for human hearts and body sizes.
What about the other federal court cases, Mitchell asks, permitting failure-to-warn claims against Taser to reach a jury based on the Tchou/Lakkireddy and Nanthakumar studies? Fontenot v. Taser International, Inc., 736 F.3d 318 (4th Cir. 2013), to start, says no such thing. That case dealt with three issues unrelated to this duty: contributory negligence, proximate cause, and excessive damages. See id. at 326. More, Fontenot involved North Carolina law, which does not limit the information the court could consider to studies published before the date of sale; indeed, it affirmatively permits such information in assessing the duty to warn. See N.C. Gen. Stat. § 99B-5(a)(2). The 2-1 decision in Fontenot also appears to be an outlier: Efforts to hold Taser liable for deaths caused by its products have not fared well in other federal courts of appeals. See, e.g., Bachtel v. TASER Int’l, Inc., 747 F.3d 965, 972 (8th Cir.2014) (holding that, under Missouri law, the plaintiff could not prove the relevant officer would have followed the allegedly necessary product warnings); Williams v. City of Cleveland, 736 F.3d 684, 687 (5th Cir.2013) (holding that, under Mississippi law, Taser’s warnings adequately conveyed the risk of serious injury or death); Marquez v. City of Phoenix, 693 F.3d 1167, 1173 (9th Cir.2012) (holding that, under Arizona law, Taser provided adequate warnings of its product’s risks); Rosa v. Taser Int’l, Inc., 684 F.3d 941, 947-48 (9th Cir.2012) (holding that, under California law, the plaintiff did not prove that Taser should have known the use of its product on suspects undergoing metabolic acidosis could result in death); Mann v. Taser Int’l, Inc., 588 F.3d 1291, 1304 (11th Cir.2009) (holding that, under Georgia law, the plaintiff could not prove use of a taser caused the suspect’s death). Nor does Piskura v. TASER Int’l, Inc., No. 1:10-cv-248, 2013 WL 1329704 (S.D.Ohio Mar. 29, 2013), conflict with our decision. That court determined that a jury could infer that an X26 caused the relevant injury based on scientific data released up to 2012, not limited to before August 2006. See id. at *6, adopting report and recom*233mendation, 2012 WL 5B78805, at *14 (Oct. 29, 2012).
One district court, it is true, has permitted a jury to infer knowledge of a material risk based on the Tchou/Lakkireddy and Nanthakumar studies alone. See Rich v. Taser lnt'l, Inc., No. 2:09-cv-02450, 2012 WL 1080281 (D.Nev. Mar. 30, 2012). It appeared to conclude that, because the studies found any possibility of cardiac capture from the X26, a jury could infer sufficient knowledge to require it to add new warnings. But the opinion neglects to mention that many prior studies, to say nothing of the many safe uses of tasers in the field, indicated the X26 would not cause cardiac problems. We for our part do not see how Taser “knew or should have known about the risk” of cardiac arrest, to quote the relevant language of the Michigan statute, based on the possible risks identified in these two studies. Mich. Comp. Laws § 600.2948(3).
What of the reality that it would have taken little for Taser to reduce any possibility (however small or unknown) of cardiac arrest by warning officers not to aim at center mass? See Dissent at 240. Michigan law does not impose failure-to-warn liability if a company does not instruct users to employ the safest possible alternative. It requires warnings only when a product’s manufacturer “knew or should have known about the risk of harm” the plaintiff alleges. Mich. Comp. Laws § .600.2948(3). That means Mitchell must offer proof that the risk of cardiac arrest actually, not possibly, exists, and she has not done so. Perhaps Michigan’s failure-to-warn regime is not the one we would have designed, but that is not for us to decide. See Erie R.R. Co. v. Tompkins, 304 U.S. 64, 78, 58 S.Ct. 817, 82 L.Ed. 1188 (1938).
The dissent’s recitation of the facts, drawn from Mitchell’s complaint, raises the question why these six police officers used a Taser at all against this 16-year-old boy. That is a fair point. It is one that explains the district court’s partial denial of summary judgment. And it is one that explains the decision of the officer defendants to settle. But it does not speak to the liability of the other defendant — Taser.
Mitchell in the end has not shown that a reasonable jury could find that Taser knew or should have known that a chest shot from an X26 posed a risk of cardiac arrest before it shipped its product to the Warren Police Department. Taser thus had no duty to warn about such a risk under Michigan law.
III.
Mitchell argues in the alternative that Taser had a post-sale duty to warn based on “subsequent studies and information that came to light after August 2006.” R. 253 at 6. Yet § 600.2948(3) offers no opening for this argument and indeed forecloses it. The statute limits the “scientific, technical, or medical information” relevant to failure-to-warn claims to information “reasonably available at the time the specific unit of the product left the control of the manufacturer.” If Taser had no such duty to warn based on the pre-sale information available, it could not be liable if later studies suggested safer ways to design and market its products. Mitchell’s post-sale duty-to-warn claim cannot succeed.
Mitchell claims that Comstock v. General Motors Corp., 358 Mich. 163, 99 N.W.2d 627 (1959), together with Michigan Compiled Laws § 600.2948(4), sidesteps this rule and allows her to bolster her negligence claim with later studies. We think not.
*234Comstock pre-dates the 1995 statute and is not to the contrary at any rate. It dealt with General Motors’ discovery in the 1950s that it had improperly manufactured hydraulic braking systems in its Buick line of cars. Despite learning of the problem shortly after the cars hit the market, General Motors did not tell its customers about the problem and how to fix it. One of the defective cars soon injured a mechanic, who sued for failure to warn of the problem. The Michigan Supreme Court held that a “duty to give prompt warning exists when a latent defect which makes the product hazardous to life becomes known to the manufacturer shortly after the product has been put on the market.” Id. at 634. That duty, Mitchell claims, applies here. The latent defect in this case, she adds, is Taser’s initial failure to warn in August 2006. See Appellant’s Br. 36-37 & n.7. Mitchell seeks to fortify this argument by pointing to Michigan Compiled Laws § 600.2948(4), which provides: “This section does not limit a manufacturer’s or seller’s duty to use reasonable care in relation to a product after the product has left the manufacturer’s or seller’s control.” As she reads the provision, the Comstock duty to warn of latent defects survived the passage of § 600.2948 and is codified in subsection 4 of the 1995 law.
Even if we assume that Comstock remains good law, it could not apply as Mitchell proposes here. Her interpretation would give § 600.2948(3) no meaningful role to play. Consider a manufacturer who had no duty to warn based on the information available at the point of sale. A year later, a scientific study comes out showing that the product was more dangerous than the manufacturer thought. If Comstock applies and if the manufacturer is now liable for a failure to warn for all sales during the first year, that liability would be based on scientific information not “reasonably available” at the point of sale. That conflicts with § 600.2948(3), which thwarts liability in just such a situation. In a battle between Comstock (a case) and subsection 3 (a statute), the statute wins. See Smith v. Martin, 124 Mich. 34, 82 N.W. 662, 662-63 (1900).
What about Mitchell’s concern that our understanding of Comstock creates the opposite problem by reading subsection 4 out of the statute? Two responses. One: It does no such thing. A manufacturer might still have a “duty to use reasonable care in relation to a product” after the point of sale if, for example, it assumed a duty to repair or service the product. See Gregory v. Cincinnati Inc., 450 Mich. 1, 538 N.W.2d 325, 335 (1995). It also might still havé a duty to warn of manufacturing defects discovered after it sold the product. See id. at 332-33. Subsection 4 thus does not lose all meaning if Comstock does not apply when the underlying defect is a failure to warn. Two: Subsection 4 does not mention Comstock, making it highly speculative to assume that it meant to codify Comstock. And to the extent any ambiguity in § 600.2948 would lead Michigan courts to consider legislative history, Lansing Mayor v. Mich. Pub. Serv. Comm’n, 470 Mich. 154, 680 N.W.2d 840, 847 (2004), the only mention of Comstock’s post-sale duty to warn there suggests that the statute overrules the duty. See Mich. Senate Fiscal Agency, S.B. 344 & H.B. 4508: Revised Enrolled Analysis 16 (1996). Subsection 3 governs this case and defeats this aspect of Mitchell’s claim.
Even if we ignored the statute and even if Comstock remained good law, Mitchell’s argument fails on its own terms. As Gregory explains, Comstock creates a duty to warn when the product in question was defective at the time of sale. See 538 N.W.2d at 331-33. But Mitchell cannot show Officer Lapham’s X26 was defective when sold under any theory of products *235liability recognized by Michigan law. She does not argue that 'Lapham’s taser suffered from a manufacturing defect — that “something [went] wrong in the manufacture process” such that Lapham’s taser was “not in its intended condition.” Prentis v. Yale Mfg. Co., 421 Mich. 670, 365 N.W.2d 176, 182 (1984). Nor does she argue that Taser defectively designed its line of X26s. See id. at 184. That leaves her only with a claim of defect based on lack of necessary warnings. As just shown, that claim fails. Comstock thus does not help Mitchell even on its own terms.
IV.
Mitchell also argues that the district court improperly dismissed her negligence claim. This claim has been a moving target from the beginning. The complaint suggests that Taser committed negligence by “continu[ing] to manufacture, market and distribute” the X26 after it knew or should have known of the risks the device posed. R. 75 at 31. In response to Ta-ser’s summary judgment motion, however, Mitchell tucked the negligence claim into a section discussing Taser’s duty to warn. There she argued that Taser assumed a duty of care through its training program and thus owed its customers a duty to warn of dangers revealed through new studies that came to light after the point of sale. The district court considered and rejected that argument. On appeal, the target has moved once again: Mitchell now claims that Taser is liable “independent of products [] liability” because it kept telling its customers to aim at center mass and kept marketing the X26 as safe after it knew or should have known better. Appellant’s Br. 44.
No matter how Mitchell pitches this claim, it cannot succeed. Start with the negligent-training angle. This theory asserts not a failure to warn but that some other affirmative aspect of the training was negligent. The Warren Police Department went through training with one of Taser’s independent contractors in September 2006, about one month after the company shipped the X26s to Warren. Even assuming that Taser could be vicariously liable, see Brinker v. Koenig Coal & Supply Co., 312 Mich. 534, 20 N.W.2d 301, 303 (1945), it owed no duty of care to Mitchell’s son Robert at that time. Under Michigan law, a person has a duty to guard only against foreseeable injuries. See Valcaniant v. Detroit Edison Co., 470 Mich. 82, 679 N.W.2d 689, 692 (2004). And the known risk of cardiac arrest was no less tentative and speculative in September 2006 than it was a month earlier at the point of sale. Taser thus had no responsibility to prevent the harm Robert suffered by refraining from instructing its customers to aim at center mass.
What about later documents where Ta-ser continued to instruct X26 users to aim at center mass? The key failing there is causation. Absent from the record is evidence that the Warren Police Department would have altered its training policies or Officer Lapham would have aimed any differently had Taser silently removed its “aim at center of mass” language from its subsequent training documents. Indeed, Mitchell does not even argue the point on appeal.
Now consider Mitchell’s theory before the district court — that Taser assumed a continuing duty to warn the Warren Police Department by virtue of its continued updating of its training materials. The Michigan Supreme Court has never held that such a duty exists. In Gregory it assumed that a continuing duty to warn can arise if “a unique or controlling relationship” develops between the manufacturer and end user. 538 N.W.2d *236at 336. But Gregory also said that no such relationship exists solely by dint of the manufacturer mailing occasional, even numerous, product updates. Id. And as far as we can tell, that was the only connection Taser had with the Warren Police Department after the initial training session. Recall, moreover, that it was not Taser who taught Warren’s training officers. That job fell to one of the company’s independent distributors. All in all, no sufficiently “unique or controlling relationship” existed to justify imposing on Taser a perpetual duty to warn Warren about all X26 dangers that surfaced after purchase.
In light of these conclusions, we need not resolve the other defenses presented by this case: (1) whether Mitchell failed to prove proximate cause because Lapham never saw the product warnings that Taser did provide; (2) whether Lapham misused the product by failing to follow Taser’s express warnings; or (3) whether Lap-ham’s violation of the Constitution and the Warren Police Department’s use-of-force policies constituted a superseding cause of Robert’s death.
For these reasons, we affirm.