This is a case about the line between fantasy and criminal intent. Although it is increasingly challenging to identify that line in the Internet age, it still exists and it must be rationally discernible in order to ensure that “a person’s inclinations and fantasies are his own and beyond the reach of the government.” Jacobson v. United States, 503 U.S. 540, 551-52, 112 S.Ct. 1535, 118 L.Ed.2d 174 (1992). We are loath to give the government the power to punish us for our thoughts and not our actions. Stanley v. Georgia, 394 U.S. 557, 565, 89 S.Ct. 1243, 22 L.Ed.2d 542 (1969). That includes the power to criminalize an individual’s expression of sexual fantasies, no matter how perverse or disturbing. Fantasizing about committing a crime, even a crime of violence against a real person whom you know, is not a crime.
This does not mean that fantasies are harmless. To the contrary, fantasies of violence against women are both a symptom of and a contributor to a culture of exploitation, a massive social harm that demeans women. Yet we must not forget that in a free and functioning society, not every harm is meant to be addressed with the federal criminal law. Because “[t]he link between fantasy and intent is too tenuous for fantasy [alone] to be probative,” United States v. Curtin, 489 F.3d 935, 961 (9th Cir.2007) (en banc) (Kleinfeld, J., concurring), and because the remaining evidence is insufficient to prove the existence of an illegal agreement or Valle’s specific intent to kidnap anyone, we affirm the district court’s judgment of acquittal on the single count of conspiracy to kidnap.
In an issue of first impression that has sharply divided our sister circuits, we must also decide the meaning of “exceeds authorized access” in section 1030(a) of the Computer Fraud and Abuse Act (“CFAA”), which imposes both criminal and civil liability. 18 U.S.C. § 1030. Specifically, we must determine whether an individual “exceeds authorized access” to a computer when, with an improper purpose, he accesses a computer to obtain or alter information that he is otherwise authorized to access, or if he “exceeds authorized access” only when he obtains or alters information that he does not have authorization to access for any purpose which is located on a computer that he is otherwise authorized to access. Because we conclude that the text, statutory history, and *512purpose of the CFAA permit both interpretations, we are required to apply the rule of lenity and adopt the latter construction. We therefore reverse the judgment of conviction as to the CFAA count.
BACKGROUND
Gilberto Valle is a native of Forest Hills, Queens. At the time of the events giving rise to his prosecution, he was an officer in the New York City Police Department living with his wife, Kathleen Mangan, and their infant daughter in Forest .Hills. Valle has no prior criminal record and there is no evidence that he ever acted violently or threateningly towards anyone.
Valle was, however, an active member of an Internet sex fetish community called Dark Fetish Network (“DFN”). He connected with individuals around the world whom he knew only by screen names such as “Moody Blues” or “Aly Kahn,” or by email addresses. Valle communicated with these individuals by email or web chat, usually in the late evening and early morning hours after his work shift. Many of his Internet communications involved the transmission of photographs of women he knew — including his wife, her colleagues from work, and some of his friends and acquaintances — to other DFN users with whom he discussed committing horrific acts of sexual violence. These “chats” consisted of gruesome and graphic descriptions of kidnapping, torturing, cooking, raping, murdering, and cannibalizing various women.
Valle’s online fantasy life was, to say the least, extremely active during this period. However, there is no evidence that he ever learned the real identities of the individuals with whom he chatted, nor is there any evidence that he ever made concrete plans to meet in person or speak by telephone or web camera with any of them.
In September 2012, Mangan became concerned about Valle’s late-night Internet activities after she found several disturbing images of dead women on a laptop that the couple shared. She installed spyware on the ■ computer, which recorded each website entered by the computer’s users and captured screen shots every five minutes. With the use of the spyware, Man-gan found more disturbing pictures and records of websites that Valle visited. These included detailed emails and chats where Valle discussed butchering her and raping and torturing other women whom they knew. After confronting Valle about his computer use and moving out of the home with their daughter, Mangan contacted federal authorities.
Valle was subsequently arrested and charged with a single conspiracy to kidnap several of the women who were the subject of his chats. Although he had chatted with numerous individuals he met on DFN, the Government identified three alleged co-conspirators: Michael VanHise, a man from New Jersey who was known to Valle as “mikevanhise81@aol.com” and “michaell9902135@yahoo.com”; an unidentified individual apparently located in Pakistan who used the screen name “Aly Khan”; and Dale Bolinger, a man in England who was known to Valle only by his screen name, “Moody Blues.” And ■ although Valle had discussed up to one hundred different women in his chats, the indictment alleged five targets of the kidnapping conspiracy: Kathleen Mangan, his wife; Alisa Friscia, Mangan’s former coworker; Andria Noble; Kristen Pontieelli; and Kimberly' Sauer, a former college classmate of Valle’s who was living in the Baltimore area.
Valle was also charged with improperly accessing a government computer and obtaining information, in violation of section 1030(a)(2)(B) of the CFAA. As an NYPD *513officer, Valle had access to the Omnixx Force Mobile (“OFM”), a computer program that allows officers to search various restricted databases, including the federal National Crime Information Center database, which contain sensitive information about individuals such as home addresses and dates of birth. It is undisputed that the NYPD’s policy, known to Valle, was that these databases could only be accessed in the course of an officer’s official duties and that accessing them for personal use violated Department rules. In May 2012, he accessed the OFM and searched for Maureen Hartigan, a woman he had known since high school and had discussed kidnapping with Aly Khan. This access with no law enforcement purpose is the basis for the CFAA charge.
The Government’s evidence at trial included the chats and emails between Valle and his alleged co-conspirators; testimony from several of the alleged targets of the kidnapping conspiracy, including his wife; other evidence seized from Valle’s computer, including videos and images he downloaded; his search term and browser history; and excerpts from a post-arrest statement. Following a 13-day trial, the jury returned a verdict of guilty on both counts. Valle subsequently moved for a judgment of acquittal pursuant to Rule 29 or, in the alternative, for a new trial pursuant to Rule 33 on both counts.
In a thorough and thoughtful 118-page opinion, the district court (Gardephe, J.) granted Valle’s Rule 29 motion with respect to the conspiracy charge. 301 F.R.D. 53 (S.D.N.Y.2014). While remaining “mindful of the jury’s critical role in our legal system,” Judge Gardephe acknowledged his responsibility to ensure that the government satisfies its burden of establishing proof beyond a reasonable doubt. Id. at 80. Emphasizing “the unique circumstances of this extraordinary ease,” he concluded that, notwithstanding the jury’s verdict to the contrary, the prosecutors had failed to prove beyond a reasonable doubt that Valle and his alleged co-conspirators had entered into a conspiracy to kidnap or that Valle had formed the requisite specific intent to kidnap. Id. at 62, 89.
In reaching this conclusion, Judge Gar-dephe cited extensively to the testimony of FBI Special Agent Corey Walsh, the lead agent assigned to. review and analyze Valle’s emails and chats whose testimony had formed (in the court’s view) the “centerpiece” of the Government’s case and the “foundation” of its argument that Valle had acted with criminal intent. Id. at 83-84. Agent Walsh testified that he, along with prosecutors and other case agents, reviewed all of the emails and chats found on Valle’s computer and concluded that Valle’s conversations with 21 of the 24 individuals whom he “met” on DFN were “fantasy.” SA 8, 128. At the same time, the prosecution team concluded that Valle’s conversations with the three alleged co-conspirators contained what they termed “elements of real crime” because they “described dates, names, and activities that you would use to conduct a real crime.” 301 F.R.D. at 65. There was no evidence that Agent Walsh or any of the other members of the prosecution, team had any specialized training or experience that would render them particularly competent to distinguish between “real” and “fantasy” chats. Indeed, Agent Walsh conceded that the “fantasy role-play” chats and emails shared many of the same features as the “real” chats and emails that purportedly reflected criminal intent, including dates for planned kidnappings, conjured acts of sexual violence, prior surveillance that Valle fantasized about having conducted, and fantastical elements such as human-sized ovens and rotisseries for cooking victims. Id. at 65-66.
*514After an exhaustive review of the chats and emails introduced at trial, Judge Gar-dephe concluded that there was no discernible difference between the “real” and “fantasy” chats:
Both sets of chats involve discussions about Facebook photographs of women Valle knows; dates for planned kidnappings; prices Valle will charge for kidnapping these women; surveillance Valle has allegedly conducted of these women; the use of chloroform to incapacitate victims; acts of sexual violence that will be perpetrated on these women; and fantastical elements such as human-size ovens arid rotisseries, and the construction of soundproofed basements and pulley apparatuses that will be used for purposes of torture.
Id. at 60. Accordingly, he concluded that no reasonable juror could have found beyond a reasonable doubt that the allegedly “real” chats evinced criminal intent any more than did the acknowledged “fantasy” chats. Id. at 84.
The district court further concluded that the Government’s remaining evidence, including Valle’s Internet search history and “real life” encounters with several of the alleged targets, was insufficient to establish either a genuine agreement to kidnap or Valle’s specific intent to kidnap in light of the fantastical nature of the chats and the weakness of the remaining evidence. Id. at 90. Judge Gardephe stressed, among other things, that there was no evidence that any of the alleged conspirators ever exchanged contact information or sought to learn each other’s true identities, and that the communications were episodic, with months often passing in between. Id. at 60. When dates for planned kidnappings came and went, Valle and his alleged co-conspirators would “simply begin discussing another woman as a potential target, in the same manner that a consumer of pornography might turn to a different image, photograph, or movie.” Id. at 89. They also had agreed to the impossible— kidnapping three different women in three different places spanning thousands of miles on the same day — and Valle had “provided his alleged co-conspirators with a veritable avalanche of false, fictitious, and fantastical information concerning himself and the steps he had allegedly taken to facilitate a kidnapping.” Id. at 61, 90. These facts, Judge Gardephe reasoned, were “entirely inconsistent with the notion that Valle was engaged in a genuine kidnapping conspiracy” and, on the other hand, “entirely consistent with Valle’s defense that he was engaged in fantasy role-play” and that the intent of the conversations was simply “mutual fantasizing.” Id. at 60, 90. Accordingly, Judge Gardephe concluded that the Government’s proof had not established Valle’s guilt beyond a reasonable doubt and granted Valle’s motion for a judgment of acquittal.
For many of the same reasons, Judge Gardephe conditionally granted Valle’s motion for a new trial on the ground that the jury’s verdict was contrary to the weight of the evidence. Id. at 104. Although the basis for his ruling was limited to the weight of the evidence, Judge Gardephe also expressed serious concern about the prosecution’s trial tactics and the effect they may have had on the jury. Specifically, he questioned the propriety of the prosecution’s repeated references to Valle’s status as a police officer, such as arguments that “it is not ok” for someone in that position to engage in such fantasies. “Once the lies and the fantastical elements [of the chats] are stripped away,” Judge Gardephe concluded, “what is left are deeply disturbing misogynistic chats and emails written by an individual obsessed with imagining women he knows suffering horrific sex-related pain, terror, and degradation.” Id. at 61. “[I]n what was an *515extraordinary case involving highly inflammatory and emotional subjects,” the prosecution’s questionable conduct had “raise[d] concerns” that the jury’s verdict was the product of “disgust and revulsion” rather than reason and that Valle had been “held to a higher standard because of his status as a police officer.” Id. at 105-07, 109.
Finally, the district court denied Valle’s motion for a judgment of acquittal as to the CFAA count. While acknowledging the existence of a “vigorous judicial debate” over the meaning of “exceeds authorized access,” the court nonetheless concluded that Valle’s conduct fell “squarely within the plain language” of the statute because Valle had not been authorized “to input a query regarding Hartigan’s name” without a law enforcement reason for doing so. Id. at 111, 113.
Valle was sentenced to 12 months in custody (which was principally a sentence of time served because he had already spent 20 months in pretrial detention), one year of supervised release, and a $25 special assessment. The Government has appealed the judgment of acquittal on the conspiracy count and Valle has appealed his conviction on the CFAA count.
STANDARD OF REVIEW
When reviewing a judgment of acquittal under Rule 29, we view the evidence in the light most favorable to the Government with all reasonable inferences resolved in the Government’s favor. United States v. Anderson, 747 F.3d 51, 60 (2d Cir.2014). The ultimate question is not whether we believe the evidence adduced at trial established the defendant’s guilt beyond a reasonable doubt, but whether any rational trier could reasonably reach that conclusion. United States v. Payton, 159 F.3d 49, 56 (2d Cir.1998) (citing Jackson v. Virginia, 443 U.S. 307, 319, 99 S.Ct. 2781, 61 L.Ed.2d 560 (1979)). Accordingly, a jury’s verdict must be upheld if any rational trier of fact could have found the essential elements of the crime had been proved beyond a reasonable doubt. United States v. Coplan, 703 F.3d 46, 62 (2d Cir.2012).
Applying this standard does not, however, mean that a reviewing court must affirm all jury verdicts. If “we are to be faithful to the constitutional requirement that no person may be convicted unless the Government has proven guilt beyond a reasonable doubt, we must take seriously our obligation to assess the record to determine ... whether a jury could reasonably find guilt beyond a reasonable doubt.” United States v. Clark, 740 F.3d 808, 811 (2d Cir.2014). This standard does not mean that if there is any evidence that arguably could support a verdict, we must affirm. In any criminal trial there is always some evidence of guilt, otherwise there could not have been a prosecution.
While we defer to a jury’s assessments with respect to credibility, conflicting testimony, and the jury’s choice of the competing inferences that can be drawn from the evidence, specious inferences are not indulged, because it would not satisfy the Constitution to have a jury determine that the defendant is probably guilty. If the evidence viewed in the light most favorable to the prosecution gives equal or nearly equal circumstantial support to a theory of guilt and a theory of innocence, then a reasonable jury must necessarily entertain a reasonable doubt.
United States v. Lorenzo, 534 F.3d 153, 159 (2d Cir.2008). We review the district court’s Rule 29 decision de novo. United States v. Reyes, 302 F.3d 48, 52-53 (2d Cir.2002).
DISCUSSION
I
To sustain a conspiracy conviction, the prosecution must prove beyond a *516reasonable doubt that the person charged with conspiracy knew of its existence and knowingly joined and participated in it. United States v. Rodriguez, 392 F.3d 539, 545 (2d Cir.2004). The Government must also prove, beyond a reasonable doubt, that the defendant possessed the specific intent to commit the offense that was the object of the conspiracy — here, kidnapping. United States v. Torres, 604 F.3d 58, 65 (2d Cir.2010). This requirement is contextual: the prosecution’s proof must be considered in relation to the rest of the evidence presented at trial, rather than in isolation. Anderson, 747 F.3d at 59.
At trial, the prosecution built its case around Valle’s chats and emails with his alleged co-conspirators. On appeal, it argues that these communications, “taken at face value, were fully sufficient to establish his intent to join a kidnapping conspiracy.” Gov’t Opening Br. 32. We disagree.
As previously explained, Valle’s chats and emails with the three alleged co-conspirators were part of a much larger set of chats and emails with 24 individuals on DFN. According to the prosecution, the former were unique because they evinced “real” criminal intent while the rest did not. After reviewing the chats and emails introduced at trial, the district court concluded that the “real” and “fantasy” chats were indistinguishable. 301 F.R.D. at 86.
Our review of the record yields the same conclusion. In both groups of chats, Valle transmits Facebook images of women and offers to kidnap and sell them on a “cash upon delivery” basis, and in both groups he expresses a desire to kidnap, rape, torture, and eat women whom he knows. In both groups Valle also claims to conduct surveillance of potential victims and discusses his intentions to kidnap them using chloroform and ropes. And in both groups he describes the various devices he “owns” that will assist in the process. Many of the “fantasy” chats also do not explicitly state that the participants are engaged in fantasy and are as graphic and detailed as the “real” chats. For example, the “real” chats and the “fantasy” chats both include haggling over the kidnapping fees that Valle “wanted to charge,” although the prosecution argues that this haggling is unique to the “real” conspiracy with Van-Hise. See id. at 84. The “real” chats thus contain the same core elements as the chats the Government concedes are “fantasy.” 1
Moreover, the “real” chats take place in the same time period as the admittedly “fantasy” chats. On the evening of July 12, 2012, for instance, Valle discusses kidnapping Andria Noble with Aly Khan in a “real” chat and, an hour later, discusses kidnapping Noble with someone else in a chat that was “fantasy.” The prosecution thus proposed that Valle simultaneously agreed to kidnap Noble while also engaging in role-play about the same woman. This temporal proximity casts further doubt upon any rational distinction between the chats.
Even when “taken at face value,” the “real” chats contain numerous other indi-cia of fantasy. For example, the prosecution alleged that Valle formed a genuine agreement with the specific intent to kidnap three different women in three differ*517ent locations on the same day. First, Valle agreed with Aly Khan to lure Man-gan to either India or Pakistan on February 20, 2012 and to slaughter her there. Second, he agreed with VanHise to kidnap Alisa Friscia in Manhattan on February 20, 2012 and deliver her to an unknown location in exchange for $4,000 in cash. Finally, Valle agreed with Aly Khan to kidnap Andria Noble on February 20, 2012 from her home in Columbus, Ohio. On appeal, the prosecution posits that the jury could have reasonably concluded that Valle seriously planned to kidnap Mangan, Fris-cia, and Noble on the same day and failed to go through with the kidnappings only because “an obstacle arose,” or because he had a “fear of getting caught.” Gov’t Opening Br. 54. We believe that no rational juror could reach this conclusion for the reason noted by Judge Gardephe: “The notion that Valle had resolved to lure Mangan to India or Pakistan [to slaughter with Aly Khan,] while at the same time kidnapping Andria Noble in Columbus, Ohio, and kidnapping Alisa Friscia from the Upper East Side of Manhattan, is simply outlandish.” 301 F.R.D. at 90.
In addition to plots that would put the same person in different places at the same time, the “real” chats are replete with references to fantastical elements such as a human-sized oven, a spit, and a remote cabin in the woods, none of which Valle owned or made any effort to acquire. The fantastical nature of the “real” chats is bolstered by the entirely virtual nature of the alleged conspirators’ relationships. Valle had no pre-existing relationship with those with whom he chatted, and he formed no real life relationship with any of them. He did not know their real names and, indeed, could not be sure of their genders, ages, or locations. Neither he nor his alleged co-conspirators made any effort to communicate by telephone, text message, or web camera, much less meet in person. And weeks or months could go by between Valle’s chats with any particular individual. While anonymity is not uncommon in Internet communications, the fantastical elements of the chats combined with the impersonal nature of the interactions provides pervasive and unmistakable indicia of deep fantasy.2
Consequently, we need look no further than the prosecution’s own work product to find reasonable doubt. The prosecution divided the exchanges into two groups and undertook to convince the jury to convict Valle on the theory that one group was fantasy and the other proved criminal intent. This exercise failed because the distinction the prosecution urged does not exist in this case. There is simply no material difference between the two groups of chats. We do not believe that the prosecution satisfies the proof beyond a reasonable doubt standard by relying upon a distinction that is untethered to reason or common sense.
Perhaps realizing that there is no actual distinction to be drawn between the “real” and “fantasy” chats, the prosecution now contends that it “did not take a position one way or the other as to whether [Valle’s online communications with people other than the named co-conspirators] constitut*518ed genuine planning, puffery, preparatory conversations, role-playing, or something else entirely.” Gov’t Opening Br. 39. The record, however, belies this assertion.
As noted, Agent Walsh was a key witness in the prosecution’s case. He was the lead investigative agent and a majority of the chats and emails introduced were admitted into evidence through his testimony. He unequivocally testified, often in response to the Government’s own questions, that the Government considered Valle’s chats with 21 other individuals to be “fantasy” and Valle’s chats with the three alleged co-conspirators to be “real.” The following exchanges between Agent Walsh and AUSA Hadassa Waxman on direct examination are illustrative:
WAXMAN: When you were reviewing those emails between [Valle] and the two dozen individuals, did you separate them into groups?
WALSH: I did.
WAXMAN: What were those groups?
WALSH: Ones that I believe that were real and ones that I believe were fantasy-
WAXMAN: Why did you make that separation?
WALSH: In the ones that I believe were fantasy, the individuals said they were fantasy. In the ones that I thought were real, people were sharing ... real details of women, names, what appeared to be photographs of the women, details of past crimes and they also said that they were for real.
WAXMAN: What caused you to make that separation between the emails you found that had realistic characteristics and those that were fantasy?
WALSH: Only my interest in obtaining information about that real criminal activity.
SA 8-9.
WAXMAN: Agent Walsh, approximately how many of Officer Valle’s emails and electronic chats did you review in connection with your investigation?
WALSH: Thousands.
WAXMAN: We just reviewed over yesterday and today about 40, is that right?
WALSH: That’s correct, ma’am.
WAXMAN: Why did you focus on these 40 particular communications?
WALSH: We believed that these chats and e-mails contained elements of real crimes.
WAXMAN: And why did you come to that conclusion?
WALSH: They described dates, names, and activities that you would use to conduct a real crime.
WAXMAN: And did you cast aside a certain number of emails as well?
WALSH: Yes, ma’am.
WAXMAN: Why did you choose not to focus on those emails?
WALSH: Quite frankly, ma’am, they didn’t seem realistic.
WAXMAN: Why not?
WALSH: They were clearly role-play. They used the word “fantasy” in the actual chats or emails.
SA 125-26. On cross-examination, Agent Walsh admitted that the grouping decision was made by numerous agents and prosecutors.
BAUM: Now, when you made that decision that 21 out of 24 participants with Mr. Valle were engaged in fantasy role-play, were you the only one who made that decision?
WALSH: No, sir.
BAUM: How many agents were involved in that decision?
WALSH: Approximately eight to 10, sir.
*519BAUM: And how many people from the U.S. Attorney’s Office were involved in that decision?
WALSH: About two, sir.
BAUM: So eight to 10 law enforcement officers and at least two lawyers from the U.S. Attorney’s Office decided that out of 24 people that Mr. Valle chatted or emailed with[,] 21 were fantasy role-plays, is that correct?
WALSH: Approximately. Yes, sir.
SA 129-30. The prosecution now urges that the distinction between “real” chats and “fantasy” role play was Valle’s defense and that the district court applied the wrong standard by forcing the prosecution to disprove the defense theory of the case. As the exchanges above demonstrate, the distinction was introduced and relied on by the Government’s case agent. In any event, intent is an essential element of the crime that the Government charged. The issue, therefore, is not whether the prosecution disproved the defense’s theory, but whether the prosecution proved its theory that Valle’s “real” chats represented a departure from his otherwise entirely imaginary world.
Alternatively, the Government argues that even if it introduced the distinction, it did not rely on or concede the truth of the distinction because it “did not even introduce any of the ‘fantasy’ conversations at trial so that [a] comparison could be made” with the “real” chats. Gov’t Opening Br. 41. A sampling of the “fantasy” chats was introduced by the defense in its cross-examination of Agent Walsh. But regardless of how the exhibits were introduced, the Government’s own investigation concluded that forty chats permitted the inference of conspiratorial intent, as compared to myriad other chats that did not. The Government claims that it does not have to prove a distinction between these two sets of chats because the jury could have rationally found that “defendants charged with attempting or conspiring to engage in criminal, deviant activity often contemporaneously engage in ‘fantasy’ behavior ... about activity ... that is similar to the charged conduct.” Id. at 43.
This contention proves too little. Once the Government constructs its case around the theory that a certain group of chats permits the inference of conspiratorial intent while another group of essentially similar chats is consistent with non-criminal behavior, some adequate explanation must be forthcoming. Where, as here, none is, the non-criminal chats are a powerful indicator that a reasonable juror must necessarily entertain reasonable doubt about the prosecution’s case.
Unable to materially distinguish the “real” chats from the “fantasy” chats, the Government relies on evidence of “real world” steps that Valle took in order to “prepare” for the kidnappings. See, e.g., Gov’t Opening Br. 56-58, 60-68. For example, the prosecution introduced evidence that Valle performed Internet searches for how to kidnap people, how to make chloroform, and how to restrain and cannibalize people. The prosecution also introduced evidence that Valle researched prior kidnappings, which it argues permitted the jury to infer that Valle was interested in how those kidnappers were caught so that he could learn from their experiences and avoid apprehension. Id. at 63.
To be sure, Internet searches can provide some relevant proof of intent. However, an Internet search, in and of itself, is not criminal. Here, the searches on which the Government relies occurred in a context of deep fantasy. As with his chats and emails, Valle’s Internet searches show that he was interested in committing acts of sexualized violence against women. Interest may be relevant evidence of in*520tent, but it does not by itself prove intent. “No doubt some people commit sex crimes because they want to turn their fantasies into reality, but most people with criminal fantasies probably refrain from acting on them, because they know it would be wrong, or because they do not want to risk the penalties.” Curtin, 489 F.3d at 962 (Kleinfeld, J., concurring).
The Government also relies on at least two occasions when Valle engaged in acts of “surveillance” of his intended victims. First, the Government notes that Valle admitted in a post-arrest statement that he was on Friscia’s block on March 1, 2012, two days after he allegedly agreed to kidnap her with VanHise. Gov’t Opening Br. 56-58. Valle told a government agent that he was on the block to drop off Mangan to have lunch with Friscia, but both Mangan and Friscia testified that they had not met for lunch that day. Valle indicated to the agent that he was on the block only very briefly, and there is no evidence to the contrary. There is also no evidence that he observed Friscia or her apartment building while he was on her block. Valle’s false exculpatory explanation for being on the block is “insufficient proof on which to convict where other evidence of guilt is weak.” United States v. Johnson, 513 F.2d 819, 824 (2d Cir.1975). As the district court found, no rational juror could conclude from this evidence alone that Valle was engaged in “surveillance.”
Second, the prosecution and our dissenting colleague contend that the jury could convict Valle of a conspiracy to kidnap based on his communications with Moody Blues about Kimberly Sauer. See Gov’t Opening Br. 35-36. This evidence is insufficient to show that Valle agreed or had the specific intent to kidnap Sauer and, in any event, it does not establish Moody Blues’s intent.
Sauer is a former college classmate of Valle’s who lives in Maryland. According to Sauer, she communicated with Valle by text message approximately ten to fifteen times a year. Mangan testified that she and Valle made three or four trips to Maryland during the course of their relationship (from 2009 through September 2012) and that each time she and Valle made an effort to see Sauer when in the area.
In January 2012, Valle asked Sauer for her address so that he could send her a Patrolmen’s Benevolent Association card. The earliest chat between Valle and Moody Blues introduced at trial takes place seven months later, on July 9, 2012. During this conversation, Valle described several girls that he was “working on grabbing ... for thanksgiving,” and told Moody Blues that “Kimberly [is] by far the easiest” to kidnap because he could “just show up at her home unannounced.” JA 80-82. After Valle suggested that “maybe you can make it here and help me with her, since you have experience,” Moody Blues responded that he lives in England but it is “easy to get to the Big apple.” JA 81.
Valle also told Moody Blues that he was “single,” had a “big gas oven,” and that “no one is around [him] for about % of a mile.” JA 84. The two then discussed how they would truss up Sauer and cook her on an outdoor spit at Valle’s mountain house. During this same chat, approximately one hour after Valle wrote that he wanted to kidnáp someone for Thanksgiving, Valle told Moody Blues that he was “thinking of a Labor Day cookout ... with Kimberly as the main course.” JA 86. Valle noted that she had “been one of my favorite victims to fantasize about for almost 10 years now.” JA 86. Again during the same chat in early July, Valle sent Moody Blues a link to a video of Sauer on *521vacation and volunteered to make chloroform and buy rope. Moody Blues replied that “Labour day is the 3rd [of] September, not a lot of time to sort out plane tickets etc. Will see what cheap deals I can get.” JA 90.
One day later, on July 10, Valle sent Moody Blues “a word document, a blueprint of everything we will need to carry this out.” JA 100. The document, entitled “Abducting and Cooking Kimberly: A Blueprint,” has a “target date” of September 2, 2012 for the abduction. It includes a photograph of Sauer, and accurately describes her age and marital status and that she is not a drug user, does not have tattoos, and drinks only occasionally. All of the other information in this document is false, including her last name, date of birth, birthplace, and educational history. The entire “plan” for abduction set out in the “Blueprint” is as follows: “I will arrive at some point Sunday night at her home to kidnap her. She lives in a quiet suburban neighborhood (Pictures of her house to be added).” The document also lists some materials that are needed, including a car, chloroform, rope, gag, tarp/plastic bags, gloves, and cheap sneakers. JA 267-68. After receiving the “Blueprint,” Moody Blues asked “[m]ay I have her address? For Googling using the Map app?” JA 101. Valle lied that he was “not sure” of her exact address. Id. There is no evidence in the record that Valle ever obtained any of the materials listed in the “Blueprint,” or that the document was ever updated with pictures of Sauer’s house or any additional information.
At some point prior to July 12, Valle called Sauer to tell her that he would be traveling to Maryland with his wife and daughter for a weekend. They made plans to meet for lunch on July 22. On July 17, Valle informed Moody Blues that he would be having lunch with Sauer. Later in this chat, Moody Blues asked Valle if he had a recipe for chloroform. Valle sent him a link. On July 19, Moody Blues again asked for Sauer’s address, and Valle replied that he did not know it by heart. Valle never provided Moody Blues with Sauer’s address.
On July 20, Valle conducted a number of Internet searches relating to kidnapping, including “how to kidnap someone,” “how to chloroform a girl,” and “kidnapped girl.” On July 21, Valle traveled to Maryland with his wife and daughter. They visited several college Mends, and had the scheduled lunch with Sauer on July 22. On July 21, Valle texted Sauer “[w]e drove by your pink building today,” and she responded “Haha yay!” JA 237. At trial, Sauer testified that she understood Valle to refer to her office building, which has pink-tinted windows, but that Valle had never visited her at work and she had never sent him photographs of the building. She described the lunch as “fine” and “pleasant.”
On the evening of July 22, after Valle returned home, he emailed Moody Blues that Sauer “looked absolutely mouthwatering.” JA 117. Valle and Moody Blues said nothing more about the plot to kidnap Sauer and did not talk again for another month. On August 21, Valle and Moody Blues began to discuss Kristen Ponticelli, a recent graduate of Valle’s high school whom he did not know. JA 264. There is no evidence in the récord that Valle and Moody Blues ever discussed Sauer or Pon-ticelli again after August 21.
As Judge Gardephe observed, the chats pertaining to Sauer are not materially different from the other fantasy chats. All of the elements of this alleged plot are equally fantastical, including the presence of the nonexistent mountain house, the human-sized oven, and the “Blueprint.” The “plan” to kidnap Sauer in the “Blueprint” is no more detailed than is the “plan” in *522Valle’s Internet chats with Moody Blues, nor does the list of materials required differ from the types of materials Valle discusses in his chats. And critically, Valle makes concerted efforts to conceal from Moody Blues any identifying information about Sauer that could be used in furtherance of a kidnapping such as her last name, date of birth, and the name of her alma mater. Although the prosecution speculates that Valle did not share accurate information about Sauer because he did not want Moody Blues to undertake the kidnapping without him, there is no evidence in the record to support such an inference.
Thus, the only meaningful difference between this alleged conspiracy and the “fantasy” chats is the occurrence of Valle’s lunch with Sauer in Maryland during approximately the same time period as he discussed kidnapping her with Moody Blues. Although the Government characterizes Valle’s communications with Sauer as “out of the blue,” the record shows that they communicated by text message in the year prior to the alleged kidnapping plot on a regular basis and that they made an effort to see each other when Valle was in town. Valle did not have lunch with Sauer alone, but rather came with Mangan and their infant daughter. Moreover, the chats between Moody Blues and Valle leading up to and following the lunch make it impossible to conclude, without speculation, that the lunch was “surveillance” in furtherance of a genuine conspiracy. Moody Blues makes only a single reference to purchasing plane tickets in the July 9 chat, but that suggestion is never brought up again. Except for the email recapping the lunch on July 22 and the August 24 conversation in which their focus moves to Ponticelli after a brief mention of Sauer, Moody Blues and Valle never again discuss Sauer or any plot to kidnap her. In fact, Moody Blues and Valle do not speak at all for the month after July 22, and the “target date” of September 2 passes with no discussion. And Valle never takes any step of any sort in furtherance of an alleged kidnapping.
We are in accord with the prosecution and our dissenting colleague that a jury might be able to distill some incriminating evidence from all of this. But “some” evidence is not the test. Because Valle’s relationship with Moody Blues is essentially indistinguishable from his relationship with all of the others with whom he chatted, we agree with Judge Gardephe that a rational jury could not conclude that this evidence was sufficient to meet the “beyond any reasonable doubt” requirement. As our case law instructs:
[I]t is not enough that the inferences in the government’s favor are permissible. A court must also be satisfied that the inferences are sufficiently supported to permit a rational juror to find that [each element of the offense] is established beyond a reasonable doubt.
If the evidence viewed in the light most favorable to the prosecution gives equal or nearly equal circumstantial support to a theory of guilt and a theory of innocence, then a reasonable jury must necessarily entertain a reasonable doubt.
United States v. Triumph Capital Grp., Inc., 544 F.3d 149, 159 (2d Cir.2008).
Finally, on the basis of this evidence, it is impossible to determine beyond a reasonable doubt whether Moody Blues — or for that matter any of Valle’s other alleged co-conspirators — ever had the specific intent to commit a kidnapping. We have taken a bilateral approach to the crime' of conspiracy: at least two people must agree. “When one of two persons merely pretends to agree, the other party, whatever he may believe, is in fact not conspiring with anyone.” See United *523States v. Bicaksiz, 194 F.3d 390, 398 (2d Cir.1999). The only evidence the Government offers to demonstrate Moody Blues’s intent is the words he used in the chats. Gov’t Reply Br. 21-22. As we have explained, these chats of “real” criminal intent are rife with indicia of fantasy and contain the same substantive elements as the chats the Government concedes are “fantasy.” The conclusion that the chats do not support a finding of Valle’s conspiratorial intent applies with equal force to Moody Blues.
On this record, no reasonable juror could conclude beyond a reasonable doubt that Valle possessed the specific intent to kidnap anyone or that he and his alleged co-conspirators ever formed an agreement to actually carry out any of the purported kidnappings. The mere indulgence of fantasy, even of the repugnant and unsettling kind here, is not, without more, criminal. We therefore affirm the district court’s judgment of acquittal as to the conspiracy count.
II
We now turn to Valle’s appeal of the judgment of conviction on the CFAA count. We reverse because section 1030(a)(2)(B) is ambiguous and where, as here, the Government and the defense both posit plausible interpretations of a criminal statute, the rule of lenity requires us to adopt the defendant’s construction. As Justice Scalia has emphasized, “[w]hen interpreting a criminal statute, we do not play the part of a mindreader.” United States v. Santos, 553 U.S. 507, 515, 128 S.Ct. 2020, 170 L.Ed.2d 912 (2008). When “a reasonable doubt persists about a statute’s intended scope even after resort to the language and structure, legislative history, and motivating policies of the statute,” Moskal v. United States, 498 U.S. 103, 108, 111 S.Ct. 461, 112 L.Ed.2d 449 (1990), we resolve doubts in favor of the defendant rather than “imputing to Congress an undeclared will” to criminalize conduct, Santos, 553 U.S. at 515, 128 S.Ct. 2020 (quoting Bell v. United States, 349 U.S. 81, 83, 75 S.Ct. 620, 99 L.Ed. 905 (1955)). The rule of lenity ensures that criminal statutes will provide fair warning of what constitutes criminal conduct, minimizes the risk of selective or arbitrary enforcement, and strikes the appropriate balance between the legislature and the court in defining criminal liability. See Yates v. United States, — U.S. -, 135 S.Ct. 1074, 1088, 191 L.Ed.2d 64 (2015); United States v. Simpson, 319 F.3d 81, 86 (2d Cir.2002).
The CFAA imposes criminal and civil liability on one who, among other things, “intentionally accesses a computer without authorization or exceeds authorized access and thereby obtains information ... from any department or agency of the United States.” 18 U.S.C. § 1030(a)(2)(B). “Without authorization” is not defined. However, “ ‘exceeds authorized access’ means to access a computer with authorization and to use such access to obtain or alter information in the computer that the accesser is not entitled so to obtain or alter.” Id. § 1030(e)(6).
The dispositive question is whether Valle “exceeded authorized access” when he used his access to OFM to conduct a search for Maureen Hartigan with no law enforcement purpose. Valle concedes that he violated the terms of his employment by putting his authorized computer access to personal use, but claims that he did not violate the statute because he never “used his access to obtain any information he was not entitled to obtain.” Valle’s Opening Br. 8. In other words, Valle argues that he did not “exceed authorized access” because he was otherwise authorized to obtain the database information about Hartigan; his *524non-law enforcement purpose in running the search is irrelevant. See id. at 9. The Government contends that Valle “exceeded authorized access” because his authorization to access OFM was limited to law enforcement purposes and he conducted a search for Hartigan with no such purpose.
The critical term — “authorization”—is not defined in the statute, but we have previously recognized in construing the CFAA that “authorization” is a word “of common usage, without any technical or ambiguous meaning.” United States v. Morris, 928 F.2d 504, 511 (2d Cir.1991). The dictionary defines “authorization” as “permission or power granted by authority.” Random House Unabridged Dictionary 139 (2001).3 Thus, common usage of “authorization” suggests that one “accesses a computer without authorization” if he accesses a computer without permission to do so at all. See, e.g., LVRC Holdings LLC v. Brekka, 581 F.3d 1127, 1133 (9th Cir.2009).
Common usage of “authorization” is less helpful in determining when one “exceeds authorized access” because it can support both Valle’s and the Government’s interpretation. While “authorization” could refer, as the Government contends, to the purposes for which one is authorized to access a computer, it could alternatively refer to the particular files or databases in the computer to which one’s authorization extends. Indeed, by its plain terms the statute is directed to improper “access.” The contested language is not “exceeds authorization,” however such authorization may be defined, but the seemingly more limited “exceeds authorized access.” 18 U.S.C. § 1030(a)(2) (emphasis added). Moreover, because “without authorization” most naturally refers to a scenario where a user lacks permission to access the computer at all, one sensible reading of the statute is that “exceeds authorized access” is complementary, referring to a scenario where a user has permission to access the computer but proceeds to “exceed” the parameters of authorized access by entering an area of the computer to which his authorization does not extend. As Judge Kozinski recognized in United States v. Nosal, “it is possible to read both prohibitions as applying to hackers: “Without authorization’ would apply to outside hackers (individuals who have no authorized access to the computer at all) and ‘exceeds authorized access’ would apply to inside hackers (individuals whose initial access to a computer is authorized but who access unauthorized information or files).” 676 F.3d 854, 858 (9th Cir.2012) (en banc).
Over the past fourteen years, six other circuits have wrestled with the question before us. Most recently, the Ninth Circuit sitting en banc in Nosal and the Fourth Circuit in WEC Carolina Energy Solutions LLC v. Miller, 687 F.3d 199 (4th Cir.2012), adopted Valle’s construction. Before that, the First, Fifth, Seventh, and Eleventh Circuits adopted the prosecution’s interpretation. See United States v. John, 597 F.3d 263 (5th Cir.2010); United States v. Rodriguez, 628 F.3d 1258 (11th Cir.2010); Int’l Airport Ctrs., L.L.C. v. Citrin, 440 F.3d 418 (7th Cir.2006); EF Cultural Travel BV v. Explorica, Inc., 274 F.3d 577 (1st Cir.2001). If this sharp division means anything, it is that the statute is readily susceptible to different interpretations.4 We therefore turn to the legisla*525tive history and motivating policies for further guidance.
Congress enacted the CFAA in 1984 to address “computer crime,” which was then principally understood as “hacking” or trespassing into computer systems or data. See H.R.Rep. No. 98-894, at 3691-92, 3695-97 (1984); S.Rep. No. 99-432, at 2480 (1986). The House Committee Report to the original bill detailed the existence of “‘hackers’ who have been able to access (trespass into) both private and public computer systems” as a result of the “corresponding proliferation of computer networking which began during the 1970’s.” H.R.Rep. No. 98-894, at 3695; see also id. at 3696 (noting the “recent flurry of electronic trespassing incidents”). The report described one instance of “computer crime” in which an individual “stole confidential software by tapping into the computer system of a previous employer from [the] defendant’s remote terminal.” Id. at 3691-92.
The Senate Committee Report to the 1986 amendments specifically described “exceeds authorized access” in terms of trespassing into computer systems or files. In heightening the mens rea requirement for section 1030(a)(2), the Committee explained that it did not want to hold liable those “who inadvertently ‘stumble into’ someone else’s computer file or computer data,” which was “particularly true in those cases where an individual is authorized to sign onto and use a particular computer, but subsequently exceeds his authorized access by mistakenly entering another computer or data file that happens to be accessible from the same terminal.” S.Rep. No. 99-432, at 2483. Congress was also careful to note that “section 1030 deals with an ‘unauthorized access’ concept of computer fraud rather than the mere use of a computer. Thus, the conduct prohibited is analogous to that of ‘breaking and entering.’ ” H.R.Rep. No. 98-894, at 3706. Consequently, the legislative history consistently characterizes the evil to be remedied — computer crime — as “trespass” into computer systems or data, and correspondingly describes “authorization” in terms of the portion of the computer’s data to which one’s access rights extend.
The Government relies upon the predecessor language to “exceeds authorized access.” As originally enacted, section 1030(a) made it a crime to “knowingly access[ ] a computer without authorization, or having accessed a computer with authorization, use[] the opportunity such access provides for purposes to which such authorization does not extend.” Counterfeit Access Device and Computer Fraud and Abuse Act of 1984, Pub.L. No. 98-473, §' 2102(a), 98 Stat. 1837, 2190 (codified as amended at 18 U.S.C. § 1030) (emphasis added). In 1986, Congress deleted the italicized language and replaced it with “exceeds authorized access.” S.Rep. No. 99-432, at 2486.
The Government argues that no substantive change was intended because the substitution was made “to simplify the language.” Id. Valle cites another provision of the Senate Committee Report, relating to subsection (a)(3), which states that Congress had “eliminate[d] coverage for authorized access that aims at ‘purposes to which such authorization does not extend,’ ” and thereby “remove[d] from the sweep of the statute one of the murkier grounds of liability under which a[n] ... employee’s access to computerized data *526might be legitimate in some circumstances, but criminal in other (not clearly distinguishable) circumstances.” Id. at 2494 (emphasis added). He argues that Congress therefore intended to abrogate any purpose-based inquiry by substituting the new “exceeds authorized access” language. While a number of courts have found this argument persuasive, see, e.g., Nosal, 676 F.3d at 858 n. 5, we have misgivings. It seems more likely that the Committee was merely explaining its removal of “exceeds authorized access” as a basis for liability under subsection (a)(3), rather than the substitution of “exceeds authorized access” in other provisions of the statute, including subsection (a)(2).
Nevertheless, we do not think that the appearance of the word “purposes” in the legislative history renders the statute clear for the simple reason that even when Congress referenced the user’s “purposes,” it spoke in terms of the particular computer files or data to which the user’s access rights extended. The Committee’s extensive discussion of subsection (a)(3) is instructive. As initially enacted, that provision made it a crime to knowingly access a government computer without authorization or exceed the scope of one’s authorization and thereby use or disclose information. S.Rep. No. 99-432, at 2494. Subsection (a)(3) therefore “swe[pt] in all computerized government information, including documents that must, under the Freedom of Information Act [ (“FOIA”) ], be disclosed to any member of the public upon proper request,” while “gloss[ing] over the reality that the existence or exact scope of a government employee’s authority to access a particular computerized data base is not always free from doubt.” Concerned that government employees would “resolve doubts against disclosure” when responding to FOIA requests, the Committee revised subsection (a)(3) in three ways, including by removing the “purposes” language. Id.
Each of' these revisions was directed toward the same problem: an employee with authorization to access certain databases entering other databases to which his authorization did not extend. And, in explaining the revisions, the Committee understood authorization in spatial terms, namely, an employee going beyond the parameters of his access rights. See e.g., id. at 2495 (declining to apply subsection (a)(3) “to access by a Federal employee of computers of that employee’s own agency,” and explaining that the revised rule “would provide prosecutors a clear, workable rule, regardless of the intricacies of a particular agency’s computer access policies: absent a fraudulent motive, an employee could not be prosecuted for simple ‘trespass’ into one of his agency’s own computers”) (emphasis added). This understanding of authorization is, as we have previously explained, consistent with Congress’s discussion of the concept elsewhere. It is likewise consistent with the statute’s principal purpose of addressing the problem of hacking, i.e., trespass into computer systems or data.
At the end of the day, we find support in the legislative history for both Valle’s and the Government’s construction of the statute. But because our review involves a criminal statute, some support is not enough. Where, as here, ordinary tools of legislative construction fail to establish that the Government’s position is unambiguously correct, we are required by the rule of lenity to adopt the interpretation that favors the defendant. Santos, 553 U.S. at 514, 128 S.Ct. 2020; United States v. Granderson, 511 U.S. 39, 54, 114 S.Ct. 1259, 127 L.Ed.2d 611 (1994). We do not think it too much to ask that Congress define criminal conduct with precision and *527clarity. As Chief Justice Marshall explained:
The rule that penal laws are to be construed strictly, is perhaps not much less old than construction itself. It is founded ... on the plain principle that the power of punishment is vested in the legislative, not in the judicial department. It is the legislature, not the Court, which is to define a crime, and ordain its punishment.
United States v. Wiltberger, 18 U.S. (5 Wheat.) 76, 95, 5 L.Ed. 37 (1820). We decline to adopt the prosecution’s construction, which would criminalize the conduct of millions of ordinary computer users and place us in the position of a legislature.
The role that the rule of lenity plays where doubt remains as to the reach of a criminal statute was discussed in Nosal, where the Ninth Circuit sitting en banc focused sharply on the same compelling concerns that Valle and amici raise on this appeal:
[T]he government’s proposed interpretation of the CFAA allows private parties to manipulate their computer-use and personnel policies so as to turn these relationships into ones policed by the criminal law. Significant notice problems arise if we allow criminal liability to turn on the vagaries of private policies that are lengthy, opaque, subject to change and seldom read. Consider the typical corporate policy that computers can be used only for business purposes. What exactly is a ‘nonbusiness purpose’? If you use the computer to check the weather report for a business trip? For the company softball game? For your vacation to Hawaii? And if minor personal uses are tolerated, how can an employee be on notice of what constitutes a violation sufficient to trigger criminal liability?
676 F.3d at 860. The Fourth Circuit, in Miller, agreed with the Ninth Circuit and echoed the same concerns:
The deficiency of a rule that revokes authorization when an employee uses his access for a purpose contrary to the employer’s interests is apparent: Such a rule would mean that any employee who checked the latest Facebook posting or sporting event scores in contravention of his employer’s use policy would be subject to the instantaneous cessation of his agency and, as a result, would be left without any authorization to access his employer’s computer systems.... [W]e do not think Congress intended ... the imposition of criminal penalties for such a frolic.
687 F.3d at 206.
We agree with the Ninth and Fourth Circuits that courts that have adopted the broader construction “looked only at the culpable behavior of the defendants before them, and failed to consider the effect on millions of ordinary citizens caused by the statute’s unitary definition of ‘exceeds authorized access.’ ” Nosal, 676 F.3d at 863; see also Miller, 687 F.3d at 206 (“[W]e believe that th[is] theory has far-reaching effects unintended by Congress.”). This is the very concern at the heart of the rule of lenity.
For example, in United States v. Kozminski 487 U.S. 931, 108 S.Ct. 2751, 101 L.Ed.2d 788 (1988), the Supreme Court refused to adopt the Government’s broad interpretation of a statute criminalizing involuntary servitude. The Government argued that the statute should criminalize “compulsion [to work] through psychological coercion as well as almost any other type of speech or conduct intentionally employed to persuade a reluctant person to work.” Id. at 949, 108 S.Ct. 2751. The Supreme Court rejected this interpretation because it would “criminalize a broad *528range of day-to-day activity,” such as “a parent who coerced an adult son or daughter into working in the family business by threatening withdrawal of affection.” Id. The Court warned that the broader statutory interpretation would “delegate to prosecutors and juries the inherently legislative task of determining what type of ... activities are so morally reprehensible that they should be punished as crimes” and would “subject individuals to the risk of arbitrary or discriminatory prosecution and conviction.” Id.
The Government does not reply substantively to Valle’s concerns about the rule of lenity or about the risk of criminalizing ordinary behavior inherent in its broad construction. It merely states that “those concerns must be raised in the first instance by individuals actually affected by the provision at issue,” and that “[t]hose cases will present fact-specific questions not relevant here, including whether the applicable authorization was clearly defined and whether the abuse of computer access was intentional.” Gov’t Opp’n Br. 15. We disagree. The Government asks that we affirm Valle’s conviction, which requires us to accept its construction of the statute. But our construction of the statute impacts many more people than Valle. It will not only affect those who improperly access information from a government computer — a result some readers might find palatable — but also those who improperly access “any protected computer” and thereby obtain information. 18 U.S.C. § 1080(a)(2)(C). As the Ninth Circuit aptly put it in Nosal, “[bjecause ‘protected computer’ is defined as a computer affected by or involved in interstate commerce — effectively all computers with Internet access — the government’s interpretation of ‘exceeds authorized access’ makes every violation of a private computer use policy a federal crime.” 676 F.3d at 859 (citing 18 U.S.C. § 1030(e)(2)(B)).
Whatever the apparent merits of imposing criminal liability may seem to be in this case, we must construe the statute knowing that our interpretation of “exceeds authorized access” will govern many other situations. See 18 U.S.C. § 1030(e)(6). It is precisely for this reason that the rule of lenity requires that Congress, not the courts or the prosecutors, must decide whether conduct is criminal. We, on the other hand, are obligated to “construe criminal statutes narrowly so that Congress will not unintentionally turn ordinary citizens into criminals.” Nosal, 676 F.3d at 863. While the Government might promise that it would not prosecute an individual for checking Facebook at work, we are not at liberty to take prosecutors at their word in such matters. A court should not uphold a highly problematic interpretation of a statute merely because the Government promises to use it responsibly. See United States v. Stevens, 559 U.S. 460, 480, 130 S.Ct. 1577, 176 L.Ed.2d 435 (2010).
CONCLUSION
For these reasons, we AFFIRM the judgment of acquittal as to the count of conspiracy to kidnap, and REVERSE the judgment of conviction as to the count of improperly accessing a computer in violation of the CFAA.
Judge STRAUB dissents in a separate opinion.
. In a "fantasy”' chat with "Tim Chase,” for example, Valle and Chase agree to kidnap a woman on January 27, 2012. Valle supplies real pictures of the woman, they agree upon a price of $4,000 for Valle’s services, and Valle states that the woman goes to the gym nightly and that he has kept a log of when she leaves and returns home. The two also agree upon a location "a hundred miles east of Erie” as the place of delivery. As Judge Gardephe found, there are no material differences between these chats with Chase, the fantasist, and Valle's chats with VanHise, the alleged co-conspirator. 301 F.R.D. at 86-87.
. As Judge Richard Posner observed in another case involving an individual engaged in sexually graphic online communications with . strangers, the defendant “may have thought (this is common in Internet relationships) that they were both enacting a fantasy.” United States v. Gladish, 536 F.3d 646, 650 (7th Cir.2008). Indeed, in Gladish the Seventh Circuit reversed a conviction for attempt to induce a minor to engage in sexual activity because "[the defendant's] talk and his sending her a video of himself masturbating (the basis of his unchallenged conviction for violating 18 U.S.C. § 1470)[we]re equally consistent with his having intended to obtain sexual satisfaction vicariously.” Id.
. See also Black’s Law Dictionary 159 (10th ed.2014) (defining "authorization” as "[o]ffi-cial permission to do something”); Webster's Third International Dictionary 146 (2002) (defining "authorization” as "the state of being authorized,” and "authorize” as "to endorse, empower, justify, permit by or as if by some recognized or proper authority”).
. The dissent claims that we "discovert] ambiguity in the statutory language where there *525is none” and summarily concludes that "exceeds authorized access” obviously encompasses a scenario where a user "did not comply with restrictions on [his] authorized access.” Dissenting Op. at 537-38. This conclusion is, with respect, not reasonable in light of these cases.