Dissenting:
I respectfully dissent because the majority’s opinion essentially finds Kleve guilty of a crime for which the jury acquitted him. Given the jury’s acquittal on the conspiracy to commit first degree murder charge and the California Supreme Court’s determination that conspiracy to commit second degree murder was not a crime at the time of Kleve’s conviction, we have no choice but to reverse Kleve’s conviction for second degree murder.
I agree with the majority that “there is no crime of conspiracy to commit second degree murder under California law.” In California, only the legislature is authorized to make conduct criminal. See People v. Mutch, 4 Cal.3d 389, 394, 93 Cal.Rptr. 721, 482 P.2d 633 (1971); People v. Smith, 57 Cal.App.4th 1470, 1480-81, 67 Cal.Rptr.2d 604 (1997); see also Bousley v. United States, 523 U.S. 614, 118 S.Ct. 1604, 1610, 140 L.Ed.2d 828 (1998) (stating that the same principle applies in the federal system). Whether conspiracy to commit second degree murder is a crime therefore must be determined by the terms of California Penal Code § 182. The mistaken dictum by the Horn court, which stated that conspiracy to commit second degree murder is a crime, did not change the nature of the law because the California legislature never intended to authorize such a crime when it enacted and amended § 182. See People v. Cortez, 18 Cal.4th 1223, 1232 n. 3, 1236, 77 Cal.Rptr.2d 733, 960 P.2d 537 (1998). Because Cortez establishes that the legislature did not provide for a crime of conspiracy to commit second degree murder at the time of Kleve’s conviction, we are not presented with the question of whether a state must retroactively apply changes in its criminal law.
As I see it, this should be the end of the analysis. Just as we held that a conviction that could have been based on a “theory of culpability that did not exist” violated due process in Suniga v. Bunnell, 998 F.2d 664, 668 (9th Cir.1993), a conviction for something that is not in fact a crime cannot be squared with the dictates of due process. Such a situation “raises the spec*1015ter of fundamental unfairness.” Id.; see also Bousley, 118 S.Ct. at 1610 (noting that habeas review can be used to correct situations in which a defendant was convicted under an erroneous interpretation of a statute and, consequently, “convicted of an act that the law does not make criminal”) (internal quotation marks and citation omitted).1
The majority, however, after acknowledging that conspiracy to commit second degree murder — the only crime for which Kleve was convicted — is not a crime under California law, embarks on some form of harmless error-like review that allows it to recast Kleve’s conviction as a conviction for conspiracy to commit first degree murder. The inescapable problem with this inquiry is that the jury acquitted Kleve of conspiracy to commit first degree murder. It is well established that the Constitution reserves the determination of guilt to the jury. See Sullivan v. Louisiana, 508 U.S. 275, 281, 113 S.Ct. 2078, 124 L.Ed.2d 182 (1993). In interpreting the jury’s second degree conviction as the “functional equivalent” of a first degree conviction, the majority erroneously supplants the jury’s clear decision to acquit Kleve of the first degree charge.
First, the jury’s verdict, when viewed in light of the jury instructions, indicates that it found insufficient evidence to convict Kleve of the first degree charge. The jury instructions explicitly distinguished between first and second degree murders and limited the circumstances under which the jury could convict Kleve of the second degree charge. We must assume that the jury, in acquitting Kleve of the first degree charge and convicting him of the second degree charge, followed its instructions. See Jones v. United States, - U.S. -, -, 119 S.Ct. 2090, 2104, - L.Ed.2d - (1999).
As Cortez indicates, the jury instructions here were internally inconsistent.2 The jury instructions explained that, in order to convict Kleve of conspiracy to commit murder, the jury had to make the threshold determination that Kleve possessed “malice aforethought.” Further, in order to convict Kleve of conspiracy, the jury had to find the “specific intent to agree to commit the public offense of murder.” The Cortez court explained that this specific intent determination constitutes a finding of premeditation and deliberation. See Cortez, 18 Cal.4th at 1232, 77 Cal.Rptr.2d 733, 960 P.2d 537.
In Kleve’s case, the jury was instructed that it next had to determine the degree of the murder. The jury was instructed that a “deliberate and premeditated killing” is “murder of the first degree.” However, the jury was instructed that, if “the evidence is insufficient to establish deliberation and premeditation,” it should deem the murder to be second degree. The instructions also specifically cautioned the jury that “if you have a reasonable doubt *1016whether such murder was of the first or of the second degree, you must give the defendant the benefit of that doubt and return a verdict fixing the murder as of the second degree.” The majority relies solely on the jury’s finding that Kleve possessed the requisite mental state for conspiracy, which, Cortez indicates, “necessarily establishes premeditation and deliberation.” Id. at 1232, 77 Cal.Rptr.2d 733, 960 P.2d 537. But the Supreme Court has “cautioned that instructions must be evaluated not in isolation but in the context of the entire charge.” Jones, - U.S. at -, 119 S.Ct. 2090, 2103. The majority thus errs in attempting to minimize the jury’s contrary finding: at the same time that the jury determined that Kleve had conspired, it necessarily found the evidence of premeditation and deliberation to be lacking.
We know from Cortez that the jury could not have believed that Kleve acted on the basis of a “mere unconsidered and rash impulse,” because the specific intent inherent in a conspiracy finding logically cannot be reconciled with the sort of rash actions that form the basis for a second degree murder conviction. See 18 Cal.4th at 1232, 77 Cal.Rptr.2d 733, 960 P.2d 537. There is also absolutely no factual basis for a conclusion that Kleve had insufficient time to premeditate because Kleve and Calley were driving around for approximately three hours. Moreover, the type of murder alleged here — a contract killing— is not susceptible to the sort of rash decision that can result in a second degree murder conviction. Therefore, based on the jury instructions, the jury must have found that there was insufficient evidence of deliberation and premeditation or, in other words, that it harbored a reasonable doubt whether the conspiracy to commit murder was of the first degree. I thus find remarkable the majority’s conclusion that Kleve actually benefitted from the jury’s “undeserved leniency” because he should have been convicted of conspiracy to commit first degree murder. The fact remains that he was not convicted of the first degree charge, and the jury instructions plainly indicate why.
The jury’s conclusion is also supported by the evidence presented at trial, which was fairly weak as to Kleve. The evidence against Kleve amounted to little more than Kleve’s presence in a car with the obviously-culpable Calley and his unlikely alibi that they were searching for an apartment for three hours. Compare United States v. Wiseman, 25 F.3d 862, 865 (9th Cir.1994) (reversing conspiracy conviction of driver of car that was loaded with marijuana, despite strong odor of marijuana in car). “The inferences arising from keeping bad company are not enough to convict a defendant for conspiracy.” United States v. Ramos-Rascon, 8 F.3d 704, 711 (9th Cir.1993); see also Wiseman, 25 F.3d at 865 (“Mere casual association with conspiring people is not enough.”). Because of Kleve’s improbable explanation, the evidence may have been barely sufficient to sustain a first degree conviction.3 Even so, it remains true that “the only evidence submitted by the government is wholly susceptible to innocent explanations.” Wiseman, 25 F.3d at 867. In light of the entirely circumstantial and not very substantial evidence, it is not surprising that the jury was unable to conclude beyond a reasonable doubt that Kleve had expressly agreed to commit first degree murder. Moreover, it is noteworthy that this was the second jury that failed to convict Kleve; the first jury deadlocked on the conspiracy to commit murder charge.
Second, I disagree with the majority’s suggestion that malice aforethought and the specific intent to conspire form a sufficient basis for a conspiracy to commit murder conviction. Cortez does not fully re*1017solve the case before us because the issue presented there was whether the trial court properly refused to give a jury instruction on second degree murder. See 18 Cal.4th at 1238, 77 Cal.Rptr.2d 733, 960 P.2d 537. The Cortez court held that the trial court acted properly because all conspiracy to commit murder is first degree murder. See id. at 1232, 77 Cal.Rptr.2d 733, 960 P.2d 537. The critical difference between this case and Cortez is that, in Cortez, there was no second degree murder instruction to create conflict with the conspiracy instruction. The Cortez majority could thus rely on the jury’s uncontra-verted specific intent finding made in the conspiracy context.
Therefore, that the Cortez court deemed sufficient jury instructions that required a finding of malice aforethought but did not specifically mention premeditation or deliberation is not dispositive. The malice aforethought instruction was adequate only because the jury had already made the necessary finding of premeditation and deliberation in adjudging the defendant guilty of conspiracy. See id. The Cortez court expressly repudiated the dissent’s argument that the court was lowering the state of mind required to convict a person of conspiracy to commit murder. “We are not concluding [that] conspiracy to commit murder ‘requires only intent to kill’4 — we are instead merely recognizing that the mental state required for conspiracy to commit express malice murder necessarily equates with and establishes the mental state of deliberate and premeditated first degree murder.” Id. at 1232 n. 3, 77 Cal.Rptr.2d 733, 960 P.2d 537 (emphasis added). Therefore, in order to affirm Kleve’s conviction we must, at a minimum, be able to say with confidence that the jury found that he had the necessary mental state of deliberation and premeditation. Yet the jury instructions, when read in their totality, vitiate any such finding. We have no way of resolving the conflict between the jury’s antithetical findings that Kleve had the specific intent necessary for conspiracy and that there was insufficient evidence or a reasonable doubt regarding whether he acted with deliberation and premeditation. As we stated in Suniga, “we cannot assume that the jury simply ignored the [erroneous] instruction.” 998 F.2d at 669.
Given this uncertainty, we definitely cannot say, beyond a reasonable doubt, that the jury found the necessary elements for a first degree conviction. Cf. Neder v. United States, - U.S. -, -, 119 S.Ct. 1827, 1834-35, 144 L.Ed.2d 35 (1999) (setting forth the standard for harmless error review).5 The equivocal tone of the majority opinion attests to this fact. Indeed, I cannot imagine what more the jury could have done to indicate that it found the evidence on the first degree charge to be lacking. Given the jury’s decision to acquit on the first degree charge and Cortez’ holding that conspiracy to commit second degree murder is not a crime, I cannot agree with the majority’s affirmance of Kleve’s second degree conviction. The majority not only usurps a proper jury function, but also relies on faulty guesswork in concluding that Kleve “could” have been convicted of the first degree charge. The Constitution “requires more than appellate speculation about a hypothetical jury’s action ... it requires an actual finding of guilt.” Sullivan, 508 U.S. at 280, 113 S.Ct. 2078.
. We reached a consistent result in another context when the Supreme Court concluded that 8 U.S.C. § 1326(b)(2) makes the existence of a prior aggravated felony conviction a sentencing factor, but that the statute does not create a separate criminal offense. See Almendarez-Torres v. United States, 523 U.S. 224, 118 S.Ct. 1219, 1226, 140 L.Ed.2d 350 (1998). Following Almendarez-Torres, we consistently reversed § 1326(b) convictions because defendants convicted of violating § 1326(b) had been convicted of something that is not a crime. See, e.g., United States v. Alviso, 152 F.3d 1195, 1199-1200 (9th Cir.1998).
. Because the error here was in the formulation of the jury instructions, we do not have a situation of alleged "inconsistent verdicts.” Kleve is not seeking to attack his second degree conviction on the basis that it cannot be reconciled with the jury's acquittal on the first degree charge, which is a very different situation that was addressed by the Supreme Court in United States v. Powell, 469 U.S. 57, 65, 105 S.Ct. 471, 83 L.Ed.2d 461 (1984). Kleve's argument for vacating the second degree acquittal has nothing whatsoever to do with the first degree charge. Simply put, he was convicted of something that is not a crime. Rather than the petitioner attempting to use the jury's acquittal on one charge to invalidate the conviction on another, here the majority uses Kleve's conviction on one charge to invalidate the acquittal on another charge.
. In light of my conclusion that we must reverse Kleve's conviction because it is for a crime that does not exist under California law, I see no need to resolve the sufficiency of the evidence question. Although we might have upheld a first degree conviction if it had occurred, because it did not in fact occur, we lack the authority to recast the second degree conviction as a first degree conviction.
. The jury instructions given in Kleve's case defined malice aforethought as "an intention unlawfully to kill a human being.”
. Of course, harmless error analysis does not apply here because the presupposition of that doctrine is that there was an actual convic-lion. As I have slated, the jury’s decision to acquit Kleve of the only valid crime for which he was charged should end the inquiry. The majority nonetheless undertakes a speculative inquiry that it does not term harmless error analysis but seems to be the "functional equivalent.”