Concurrence by Judge Kozinski
OPINION
BERG, District Judge:In 2015, Christopher Preston was convicted on two counts of aggravated sexual abuse of a child. He appeals, arguing that evidentiary errors and prosecutorial misconduct rendered his trial fundamentally unfair. We agree. There were a number of trial errors and, considering that evidence of guilt was not overwhelming, their cumulative effect prejudiced Preston. Accordingly, we REVERSE.
I. BACKGROUND
A. Statement of Facts
In 1998, Christopher Preston lived with his then-wife Andrea Preston on the Toho-no O’odham reservation in Tucson, Arizona, where he worked as an electrician. Preston befriended one of his colleagues, Sean Fox, who had three stepsons—Timothy, Barry, and Mitchell Rosenberg. Mr. Fox, his stepsons, and his wife Kathleen (the boys’ mother), would occasionally visit the Prestons’ home to socialize. Other times, Mr. Fox went over with just the boys to strip copper or play catch.
That year, Preston was an assistant coach for a little league baseball team in northwest Tucson. He arranged for Timothy Rosenberg (“Rosenberg”)—the alleged victim in this case, who was ten at the time—to join the team. The fields the team practiced and played on were about an hour’s drive from Sean Fox and Kathleen Rosenberg’s home. Because of this logistical challenge, Rosenberg’s parents generally did not transport him to his games or practices. Instead, Preston did.
Some weekends, the team played in tournaments spanning Friday, Saturday night, and Sunday. On such weekends, Rosenberg would stay overnight at Preston’s home on both Friday and Saturday. Rosenberg’s brother Barry, who was fourteen at the time, testified that these overnights occurred on ten or more occasions. When the 1998 Little League season ended, Rosenberg stopped going to Preston’s house. A year or two later, the Fox/Rosenberg family moved to Kansas and lost touch with the Prestons.
By 2012, Rosenberg was a twenty-four-year-old living in Kansas, experiencing troubles with the law and abusing drugs and alcohol. On March 1-2 that year, he was admitted to an emergency room in Wichita for an anxiety attack. Upon discharge from' the hospital, he went to his mother’s home. In talking with his mother, Rosenberg disclosed that Preston had molested him in 1998. This was the first time Rosenberg had revealed this information to anyone. Ms. Rosenberg called the police and arranged for her son to see her former therapist, Gail Bussart.
Bussart treated Rosenberg from March 2012 to January 2013. During treatment, Rosenberg told Bussart that Preston sexually abused him over an eighteen-month period beginning when he was ten. He did not, however, provide details. Bussart stopped seeing Rosenberg on January 3, 2013, because she thought he was lying about his substance abuse.
From late March through late April 2013, Tohono O’odham officers and FBÍ agents interviewed Rosenberg. Unlike in his conversations with Bussart, Rosenberg provided them with details about the alleged molestation. Specifically, he told a Tohono O’odham officer that Preston molested him when he was seven or eight years old and that he clearly remembered it happening twenty times. He added that the abuse occurred on Preston’s living-room couch and that Preston would put his penis between Rosenberg’s legs and direct Rosenberg to masturbate him until he ejaculated. In addition, Rosenberg met with two FBI agents and, before the meeting, sent them- a journal that he had kept throughout his treatment fry Bussart (although, according to Rosenberg, Bussart never read it). In the journal, Rosenberg recorded previously unrevealed information, including that: Preston and Rosenberg had fellated one another; Rosenberg was not certain whether he had been anally penetrated; and Rosenfrerg once saw Preston in his room watching pornography with a bottle of lubricant.
In October 2012, Tohono O’odham Detective Manny Rodriguez interviewed Preston about Rosenberg’s allegations, which Preston denied. The interview was recorded. A few days later, FBI Special Agent Mark Dellacroce interviewed Preston and administered a polygraph examination to him. This interview was not recorded. Dellacroce testified at trial that, during the interview, Preston denied Rosenberg’s, .allegations, but also stated that he “could not remember” receiving oral sex from Rosenberg because at that time “[Preston] was a meth addict.”
B. Procedural History
On October 23, 2013 a grand jury returned an indictment charging Preston with two counts (Counts 1 and 2) of aggravated sexual abuse of a child, in violation of 18 U.S.C. § 2241(c), and two counts (Counts 3 and 4) of abusive sexual contact of a child, in violation of 18 U.S.C. § 2244(a)(5).
On August 10, 2015, a six-day jury trial commenced. The only direct evidence offered at trial was Rosenberg’s testimony. Although Preston did not testify, he presented evidence of his denials through the testimony of the law enforcement officials who interviewed him. At the close of its case, the government conceded that the evidence did not support a conviction on Count 4, which was dismissed. On August 18, 2015, the jury found Preston guilty of Counts 1 and 2 and not guilty of Count 3. On October 26, 2015, the district court sentenced Preston, to concurrent terms of 162 months in prison on Counts 1 and 2 arid imposed concurrent terms of lifetime supervised release and special assessments totaling $200.
On appeal, Preston argues that the district cdurt and the prosecutor committed a variety of errors and that these errors— either independently or cumulatively—deprived him of his, right to a fair trial. The testifying witnesses relevant to his appeal include Gail Bussart (Rosenberg’s therapist), Agent Dellacroce (the FBI agent who interviewed Preston), Andrea Preston (Preston’s ex-wife), Timothy Rosenberg (the alleged victim), Barry Rosenberg (Rosenberg’s brother), and Dr. Simpson (Preston’s memory expert).
II. STANDARD OF REVIEW
This Court reviews challenged evidentiary rulings for abuse of discretion, United States v. Torralba-Mendia, 784 F.3d 652, 659 (9th Cir. 2015), and, if the district court erred, usually then asks whether the error was harmless, United States v. Job, 851 F.3d 889, 902 (9th Cir. 2017). Claims of prosecutorial misconduct are also generally reviewed under the harmless error standard. United States v. Alcantara-Castillo, 788 F.3d 1186, 1190 (9th Cir. 2015). Where a defendant raises an issue on appeal that was not raised before the district court, the review is for plain error. United States v. Pelisamen, 641 F.3d 399, 404 (9th Cir. 2011).
Where, however, as here, there are multiple trial errors, “‘a balkanized, issue-by-issue ... review1 is far less effective than analyzing the overall effect of the errors in the context of the evidence introduced at trial against the defendant.” United States v. Frederick, 78 F.3d 1370, 1381 (9th Cir. 1996) (quoting United States v. Wallace, 848 F.2d 1464, 1476 (9th Cir. 1988)). This is because the cumulative effect of multiple trial errors “ ‘can violate due process even where no single error ... would independently warrant reversal.’’’ Parle v. Runnels, 505 F.3d 922, 927 (9th Cir. 2007) (citation omitted); see also, e.g., Thomas v. Hubbard, 273 F.3d 1164, 1181 (9th Cir. 2001). In deciding whether the combined effect of multiple errors prejudiced a defendant we ask whether the errors stand in “ ‘unique symmetry ,.., such that [they] amplify each other in relation to a key contested issue in the case.’ ” Ybarra v. McDaniel, 656 F.3d 984, 1001 (9th Cir. 2011) (quoting Parle, F.3d 505 at 933).
III. DISCUSSION
Preston raises over fifteen individual trial errors, across seven different categories. We reverse based on the cumulative effect of the following: (1) improper witness testimony that bolstered Rosenberg’s credibility and offered opinion on the credibility of sex abuse allegations in general; (2) prejudicial propensity evidence in the form of Preston’s ex-wife’s testimony regarding a child-incest fantasy Preston allegedly had in 2003; and (3) prosecutorial misconduct, namely: commenting on Preston’s decision not to testify; witness vouching, and misstating the evidence in summation. Because we find cumulative error, we do not decide the prejudice caused by any of these individual errors, nor do we reach the merits of the remaining errors Preston alleges.
A. Testimony Bolstering Rosenberg and Opining on Sex Abuse Generally
The first set of trial errors we discuss arises from testimony by Gail Bussart, Barry Rosenberg, and Agent Dellacroce that suggested Rosenberg’s allegations of abuse were believable or were likely to be true. This set of errors also involves a portion of Bussart’s testimony, offered as lay opinion, which opined.on the general believability of sex abuse allegations and on whether Rosenberg demonstrated emotions consistent with sex abuse - victims generally.
1. Legal Standards
Just as “[i]t is emphatically the province and duty of the judicial department to say what the law is,” Marbury v. Madison, 5 U.S. (1 Cranch) 137, 177, 2 L.Ed. 60 (1803), it is emphatically the “province and duty [of the jury] to determine ... the weight and the credibility of the testimony of the witnesses .... ” Allis v. United States, 155 U.S. 117, 121, 15 S.Ct. 36, 39 L.Ed. 91 (1894); United States v. Bonds, 784 F.3d 582, 603 (9th Cir. 2015) (“[W]e must respect the exclusive province of the jury to determine the credibility of witnesses ....”) (citation omitted). Accordingly, “testimony regarding a witness’s credibility is prohibited unless it is admissible as character evidence.” United States v. Sanchez-Lima, 161 F.3d 545, 548 (9th Cir. 1998).
Additionally, while expert witnesses may testify in the form of opinion as to general matters based on specialized knowledge, Fed. R. Evid. 702, lay witnesses may not. Fed. R. Evid. 701.
2. Gail Bussart’s Testimony
The first portion of Bussart’s testimony alleged to be error arises from an email that she wrote to one of her supervisors, stating, “[I] saw [Rosenberg] on January 3, 2013, I suspected lies and dishonest behavior at that time ... I reinterated [sic] to [Rosenberg] the necessity of clean and sober behavioral [sic] in order for this therapist to continue with services.” In a written opinion overruling the government’s objection, the district court admitted this email, under Fed. R. Evid. 608, as Bussart’s opinion of Rosenberg’s character for untruthfulness with respect to drug and alcohol use. The Court added, “[t]he Government, of course, can cross-examine the therapist on her opinion as to whether her opinion is limited to lies about drug use or is made more broadly.”
Preston argues—and we agree—that there were three instances of error related to this email during Bussart’s testimony:
First, on direct examination, the government and Bussart had the following exchange:
“Q: [When you emailed your supervisor] that you thought Tim was lying to you ... did you think he was lying to you about the alcohol and drugs or did you think he was lying to you about the sexual abuse?
A: The alcohol and drug use.
Q: Not the sexual abuse?
A: Not the sexual abuse.”
Defense counsel objected to the government’s initial question on the bases of foundation and speculation, but the district court overruled the objection.
Second, on cross-examination, defense counsel and Bussart had the following exchange:
“Q: [A]s far as the suspicion of lies, you just say, “I suspected lies, dishonest behavior at the time,” but you don’t specifically reference or say about drug use?
A: I only talked to her about his alcohol and drug use, nothing else [... ]
Q: And you can’t say for sure what else he has possibly lied to you about?
A: I can say he wasn’t lying about his sexual abuse. It is my opinion, my professional opinion.
Third, as its final question on redirect, the government asked Bussart, “[i]s it your opinion that Tim Rosenberg was truthful to you about the sexual abuse?” She replied, “[i]t is, yes.”
Shortly after this question was asked, defense counsel objected and the district court overruled the objection, stating that it was “tardy” and that the question was “appropriate ... in view of the questions that were asked in cross.” The next day, however, the district court issued a curative instruction to the jury regarding Bus-sart’s testimony. The court stated, “You are the exclusive judges of who to believe ... so to the extent that Ms. Bussart offered any opinion or belief about the truthfulness of another witness, you must disregard that and decide for yourselves .... ” In its final charge to the jury, the court repeated this instruction.
The government does not contest that it was improper for Bussart to indicate that she believed Rosenberg was telling the truth about the sexual abuse allegations. Rather, the government argues that we should not review this issue because, in moving to admit Bussart’s email, the defense invited any error that occurred. The government points out that the district court’s order admitting the email “expressly allowed the [prosecutor] to clarify the extent of [Bussart’s] opinion.”
The defense did not invite this error. The district court allowed the defense to question Bussart about her opinion in the email regárding Rosenberg’s truthfulness about drug and alcohol use and granted the government permission to “cross-examine [her] ... as to whether her opinion is limited to lies about drug use or is made more broadly.” This ruling did not authorize the government to elicit Bussart’s opinion—-which the prosecutor did on both direct and redirect—as to whether Rosenberg was telling the truth about the ultimate issue: whether Preston sexually abused him.1 Although some of defense counsel’s cross-examination may have expanded the area of inquiry (for example, “you can’t say for sure what else he has possibly lied to you about?”), this line of questioning was a response to the government’s elicitation of Bussart’s statement that she believed Rosenberg’s allegations. Defense counsel did not invite Bussart’s three clearly improper statements regarding her belief in the veracity of Rosenberg’s allegations, and the district court abused its discretion by allowing them.
In addition to improper opinion testimony indicating that she believed Rosenberg individually, Bussart gave improper opinion testimony as a lay witness about whether sex abuse victims generally tell the truth. The government never noticed Bussart as an expert, and the district court denied the government’s request, the day before trial, to offer expert testimony concerning a worksheet entitled “Adult Grooming Behaviors for Sexual Abuse of Children.” The court instructed the government that Bussart was allowed to testify about her treatment of Rosenberg specifically and about grooming, but only to the extent that the topic came up during treatment. The court did not allow her to offer an opinion as an expert on any topic.
At trial, Bussart offered two opinions regarding sexual abuse generally, and the district court erred in allowing her to do so.
First, although defense counsel failed to object, the district court plainly erred in allowing Bussart to answer the following juror question: “To the best of your knowledge, have you treated males for sexual abuse only to find the abuse to be false?” Bussart replied, “I have not experienced that even when I have done—treated men and women or girls or boys, yes, that the allegation has normally been true.” The court should have stricken this testimony. Testifying, as a lay witness, Bussart did not offer any factual basis in support of the assertion that all the allegations made by people she treated had “normally been true.”
Second, the district court abused its discretion by permitting the following juror question over: defense counsel’s objection, which was posed by a juror to Bussart: “Based off of your sessions with Tim and his emotions, did his emotions demonstrate similarity to those of a person that had been a victim of sexual abuse?” Bussart replied, “Yes. Very much so.” The court had not qualified Bussart to offer expert opinion concerning how1, sex abuse victims generally demonstrate their emotions. To permit this question and response was an abuse of discretion, running afoul of the court’s instruction that Bussart could only' testify about her personal experience in treating Rosenberg.
In "sum, the district court abused its discretion in (1) allowing the government’s line of questioning that led to Bussart’s three improper statements indicating that she believed Rosénberg’s allegations, and (2) permitting a juror’s question to be asked regarding whether Rosenberg demonstrated his emotions in a manner consistent with sex abuse victims generally. In addition, it was plainly erroneous for the district Court" to allow Bussart to state that allegations of sexual abuse'in her'patients had normally been true. We do not decide the prejudicial effect of these errors independently. Instead, we reserve our analysis of the prejudicial effect of Bussart’s testimony until the end of'the opinion, where wé discuss "the cumulative effect of all the errors at trial. See infra Part III.D.
3. Barry Rosenberg’s Testimony
The next improper opinion testimony we consider is that offered by Barry Rosenberg, Tim Rosenberg’s brother.
In the summer of 1998, Barry, then fourteen years old, also frequently spent time at Preston’s home. At trial he testified about his experiences there. Preston contends that the following testimony— elicited in response to a juror question— impermissibly allowed Barry to offer his opinion that he found his brother’s allegations of sexual abuse to be believable:
Q: “What was your response when you learned about the allegations of sexual abuse by the defendant?”
A: “[T]here wasn’t any questioning my brother about what he was saying. There was just understanding. I can understand. I can see. Things make sense to me.”
The defense did not object to this question or to Barry’s answer, and the issue was therefore not preserved for appeal. Consequently, we review for plain error. The government argues that the statement “there wasn’t any questioning my brother about what he was saying” was ambiguous. While that statement may be interpreted in different ways, one reasonable way of interpreting it is that Barry found “what [his brother] was saying” to be beyond question. In any-event, -Barry’s four subsequent statements—“There was "just understanding. I can understand. I can see. Things make sense to me”—-were more definitive. In this context, the “things” that Barry understood,- saw, and made sense to him, were his brother’s allegations that Preston sexually abused him. Because a reasonable juror would have understood this testimony as Barry professing his belief in the veracity of his-brother’s allegations, permitting this testimony was plain error. Barry Rosenberg’s statements will factor into our -cumulative error analysis.
k. ■ Agent-Dellacroce’s Testimony
In the course of his investigation, Special Agent Dellaeroce interviewed Preston and gave him a polygraph exam. In a pretrial ruling, the district court held that Dellacroce could not mention the polygraph or its results but could testify about the questions he asked Preston during the exam, as well as Preston’s answers. During direct examination, the government and Dellacroce had the following exchange:
Q: [...] Did you ask the defendant did you put your penis in [Rosenberg’s] mouth at your house?
A: Yes.
Q: And what was the defendant’s response to [the] question[ ]?
A: No.
Q: Did you indicate to the defendant that you did not believe him?
A. Yes.
Q: What was the defendant’s response when you told him you didn’t believe him?
A: Preston never denied receiving blow jobs from. [Rosenberg] ... [he] stated that he could not remember receiving blow jobs from [Rosenberg] because that is when he was a meth addict.”
Preston made no objection about this testimony, and the government, citing an unpublished disposition, argues that it was not plain error to allow it because Della-croce was merely describing an interrogation technique—he told Preston he did not believe him in order to try and elicit a confession. The record before us does not support the government’s position that Agent Dellacroce was testifying to his in.vestigative techniques—or that the jury would have any way of knowing that. We simply have the question, “Did you indicate to the defendant that you did not believe him?” and the answer, ‘Yes”—a statement from a witness that the defendant’s denial was not credible. Moreover, no curative instruction was given. It was thus plainly erroneous for the district court to allow Dellacroce to testify that he did not believe Preston’s.denial of Rosenberg’s allegations. We will weigh Della-croce’s statements in our cumulative error analysis.
B. Preston’s Alleged 2003 Child-Incest Fantasy
Preston contends that the district court erred in admitting the-testimony of his ex-wife—Andrea Preston—that in 2003, five years after Preston allegedly abused Rosenberg, she witnessed Preston masturbating while looking at a computer image of his eight-year-old stepson clothed in socks and underwear.
Before trial, defense counsel moved to exclude this testimony as impermissible propensity evidence and the government responded that it should be admitted -under Fed. R. Evid. 404(b)- to prove Preston’s intent to molest Rosenberg. In a written order, the-trial court denied the motion and found the evidence to be admissible under the intent exception to Rule 404(b). At trial, as expected, Andrea Preston testified to having witnessed the defendant masturbating to a picture of his eight-year-old stepson some five years after the abuse charged in the indictment allegedly occurred. In its summation, the government reiterated her testimony;-adding that this evidence, “only goes to [Preston’s] intent and that’s it.”
For the reasons set forth below, we find that the district court abused-its discretion in-allowing Ms. Preston’s testimony,
1. Legal Standards
Rule 404 of the Federal Rules of Evidence creates a general prohibition on using evidence of a person’s- character or character trait in order to prove that on a particular occasion a person acted in accordance with that character or trait. Fed. R. Evid. 404(a)(1). There is also a.particular prohibition against using evidence of other crimes, wrongs, or acts to prove a person’s character in order to show that on a particular occasion the person acted in accordance with that character. Fed. R. Evid. 404(b)(1). This same Rule, however, carves out certain “permitted uses” of such evidence. Specifically, evidence of other crimes, wrongs, or acts may be admissible “for another purpose, such as proving motive, opportunity, intent, preparation, plan, knowledge, identity, absence of mistake, or lack of accident.” Fed. R. Evid. 404(b)(2). Even where 404(b) evidence falls within a permitted purpose, it should be excluded, under Rule 403, if the court finds that its probative value is substantially outweighed by a danger of unfair prejudice. Fed. R. Evid. 403.
This case involves evidence of an “other act” that occurred some five years after the events charged in the indictment. We have held that evidence of a subsequent act may be admitted under 404(b). United States v. Lloyd, 807 F.3d 1128 (9th Cir. 2015). But in introducing other act evidence, the government always must show: (1) that the act tends to prove a material element or point, (2) that the act is not too remote in time from the crime charged, and (3) that the evidence is sufficient to support a finding that the defendant committed the subsequent act. Id.
When seeking to introduce other act 404(b) evidence for the purpose of proving intent, the proposing party must show that the other “act is similar to the offense charged.” United States v. Hadley, 918 F.2d 848, 851 (9th Cir. 1990) (evidence of prior isolation and sodomy of students sufficiently similar to charged crime of isolation and sodomy of a student); see also, e.g., United States v. Vo, 413 F.3d 1010, 1018-19 (9th Cir. 2005) (prior conviction for cocaine trafficking sufficiently similar to charged crime of methamphetamine distribution); United States v. Ayers, 924 F.2d 1468, 1474 (9th Cir. 1991) (prior purchase of numerous cashier’s checks, for amounts that did not trigger banks’ duty to report them to the Internal Revenue Service, sufficiently similar to charged crime of tax evasion). This showing of similarity is necessary because if the other act is not sufficiently similar to the crime charged, “it does not tell the jury anything about what the defendant intended ... unless, of course, one argues (impermissi-bly) that the [other] act establishes that the defendant has criminal propensities.” United States v. Miller, 874 F.2d 1255, 1269 (9th Cir. 1989).
2. Analysis
In a written Order admitting Ms. Preston’s testimony about the 2003 masturbation incident, the district court held in relevant part:
“Evidence that the Defendant was sexually interested in a young boy approximately the same age the victim was at the time of the charged events could make it more likely that the Defendant had the intent to commit sexual acts in this case or that such acts were not a mistake. The evidence is, therefore, relevant. The similarity between the victim’s age at the time of the charged acts and the age of the boy in the image creates a high probative value for that evidence, and it is not substantially outweighed by any prejudice caused to the Defendant. Finally, the evidence is sufficiently similar under Rule 404(b) to show the Defendant’s intent because of the similarity in the age of the victim and boy in the image; approximately five years subsequent is not too remote in time, see United States v. Johnson, 132 F.3d 1279, 1283 (9th Cir. 1997); eyewitness testimony from the Defendant’s ex-wife is sufficient to introduce the evidence to the jury; and, the evidence is probative- of the Defendant’s intent to engage in sexual conduct, an essential element of at least one charged act.”
On appeal, Preston argues that the subsequent act was not sufficiently similar to the crime charged to be properly admitted via 404(b) under an intent theory of admissibility. Moreover, Preston contends that the district court should have excluded the evidence under Rule 403, because intent was not in dispute in this case and therefore any marginal probative value of this evidence was substantially outweighed by the danger that the jury would draw a propensity inference against Preston on the basis of his ex-wife’s highly prejudicial testimony.
We agree with Preston. In admitting Andrea Preston’s testimony, the district court abused its discretion in two ways. First, under Hadley, the district court must find that the “other act” sought to be introduced under 404(b) to prove intent is similar to the crime charged. Here, the district court found only that the age- .of Preston’s stepson in the photograph to which Preston masturbated (8 years old) was similar to Rosenberg’s age when Preston allegedly molested him (10 years old). This finding does not adequately explain or discuss how the act of masturbating to a picture of a boy in underwear—a noncriminal act—is similar to the crime of real-life sexual abuse of a child. The district court therefore abused its discretion in admitting Ms. Preston’s testimony without complying with Rule 404(b)’s similarity requirement expressed in Hadley and other cases.
Second, the district court abused its discretion by finding the evidence admissible under Rule 403. Rule 403 is meant to “ensure that potentially devastating evidence of little probative value will not reach the jury.” United States v. LeMay, 260 F.3d 1018, 1026 (9th Cir. 2001). We have long héld that “[w]here the evidence is of very slight (if any) probative value, it’s an abuse of discretion to admit it if there’s even a modest likelihood of unfair prejudice or a small risk 'of misleading the jury.” United States v. Wiggan, 700 F.3d 1204, 1213 (9th Cir. 2012) (quoting United States v. Hitt, 981 F.2d 422, 424 (9th Cir. 1992)).
Evidence of a subsequent masturbation incident linked to child-incest fantasy is highly prejudicial to a defendant charged with child molestation; a jury confronted with such disgusting evidence is likely to conclude that the defendant “is the type” to molest a child. See United States v. Curtin, 489 F.3d 935, 964 (9th Cir. 2007) (Kleinfeld, J. concurring) (“incest has had a rare power to disgust” (quoting Richard A. Posner, Sex and Reason 201 (1994)). Moreover, as this Court has recognized, in many cases, the “link between fantasy and intent is too tenuous to be probative,” as “[p]eople commonly fantasize about doing things they have no intention of actually doing.”2 And fantasy is even less probative of intent in cases where, as here, intent is not actually disputed—that is, where the defense is a general denial of committing the offense, rather than an admission to an act coupled with a specific denial of the requisite intent. When the defendant generally denies committing the offense of sexual abuse, the victim’s testimony that the abuse occurred—if accepted by the jury—effectively proves all of the elements of the offense, including that the defendant intended to seek sexual, gratification. In such circumstances, proof'of another act showing a defendant’s sexual proclivities toward children contributes little to the government’s case. It just tempts the jury to draw the impermissible inference that the defendant has a propensity to sexually abuse children.3
Here, the “visceral impact” of. this evidence “far exceeds [its] probative value.” See Hitt, 981 F.2d at 424 (finding that mere “photographs of firearms often have a visceral impact that far exceeds their probative value”). Because Preston’s alleged 2003 incest fantasy had little probative value (abuse and fantasy are dissimilar acts, and Preston did not put intent at issue in the case), and great potential to inspire prejudice (incest is a uniquely disgusting act, and juries are apt to draw a propensity inference from other act evidence), the district court abused its discretion in admitting it.
In sum, the district court abused its discretion in (1) admitting Andrea Preston’s testimony under Rule 404(b) without first establishing how Preston’s masturbation fantasy was similar to sexually abusing Rosenberg, and (2) admitting the evidence under Rule 403 given its slight probative value and high risk of unfair prejudice. We will analyze this prejudicial effect within our cumulative error analysis. ’
C. Prosecutorial Misconduct
:The next set of alleged trial errors relates to prosecutorial misconduct. Specifically, Preston argues that the prosecutor improperly commented on Preston’s decision not to testify, impermissibly vouched for Rosenberg, and misstated the evidence in summation.
1. Commenting on Preston’s Failure to Testify
To realize the Fifth Amendment’s guarantee against compelled self-incrimination, the Supreme Court has held that it is improper for a prosecutor to comment on a defendant’s decision not to testify. Griffin v. California, 380 U.S. 609, 615, 85 S.Ct. 1229, 14 L.Ed.2d 106 (1965). The government may comment on the defense’s argument, but may not make a comment “ ‘if it is manifestly intended to call attention to the defendant’s failure to testify, or is of such a character that the jury would naturally and necessarily take it to be a comment on the failure to testify.’ ” Rhoades v. Henry, 598 F.3d 495, 510 (9th Cir. 2010) (quoting Lincoln v. Sunn, 807 F.2d 805, 809) (9th Cir. 1987)). Thais, where a defendant is the only possible witness who could rebut the testimony of the government witnesses, it is inappropriate for a prosecutor to point out the lack of witnesses or testimony on the other side, because “ ‘this can only cause the jury to naturally look to the only other evidence there is—the defendant—and, hence, this could be a prohibited comment on the defendant’s failure to testify.’” Lincoln at 809 (quoting United States v. Cianciulli, 482 F.Supp. 585, 591-92 (E.D. Pa. 1979)).
Preston argues that the prosecutor impermissibly commented on his failure to testify when the government stated in summation, “[Rosenberg] was sexually abused. He told you under oath on the stand. There’s no evidence, there’s no testimony in this case that contradicts Tim Rosenberg’s testimony,”
Here, although the prosecutor did not mention Preston by name, the government’s comment was impermissible under Lincoln because it directed the jury’s attention to a lack of testimony contradicting Rosenberg. The only person who could have provided testimony to contradict Rosenberg’s version of the events was Preston, since the only other two persons present in the home at the time the abuse allegedly occurred—-Barry Rosenberg and Andrea Preston—testified that' they never suspected anything. In this context, it was plain error for the prosecutor to state that “there’s no testimony in this case that contradicts Tim Rosenberg’s testimony,” because the jury would have immediately inferred that they did not hear testimony from Preston, the only witness who could have directly contradicted Rosenberg’s allegations.4 We reserve an analysis of the prejudicial effect of the prosecutor’s comment for our cumulative error analysis.
2. Vouching for Rosenberg
Preston further argues that the government impermissibly vouched for the veracity of Rosenberg’s testimony. “Improper vouching consists of placing the prestige of the government behind a witness through personal assurances of the witness’s veracity, or suggesting that information not presented to the jury supports. the witness’s testimony.” United States v. Ruiz, 710 F.3d 1077, 1085 (9th Cir. 2013) (citation omitted).
Preston contends that the government offered personal assurances of Rosenberg’s veracity in the following instances:
1. In opening, the government stated, “The truth always comes out no matter how many years have passed ... And the truth in this case is that the defendant, Christopher Preston, put his penis in ten-year-old Timothy Rosenberg’s mouth .... ”
2. In summation, the government reiterated, “The truth will always come out .... And in this case the truth has come out.” The government further noted, “In the end ... ladies and gentlemen, the truth came out of the mouth of Tim Rosenberg.”
3. In rebuttal, referring to Rosenberg’s testimony, the government stated, “And that is the truth in this case. That is the evidence in this case.” Later in rebuttal, the government added, “[Rosenberg] has come here to court ... to testify to you as to the truth of what happened ... the uncontroverted truth as to what happened.” The government also stated, “the crux of this case, the core of the truth of what Timothy Rosenberg testified to in detail on the stand under oath is that the defendant sexually molested him as a child. ...” Finally, the government declared, “[Rosenberg] probably wishes that the drinking and the drugs ... could erase this memory of what happened. But it can’t. And it didn’t. Because it is the truth of what happened.”
The defense objected only to the final comment. Although the court overruled the objection, the government did remind the jury that “what attorneys argue to you obviously is not evidence” and stated “we are not vouching for [Rosenberg] ... what the government is saying is this is what the evidence shows, this is what the testimony shows, this is what—Tim’s testimony has been corroborated.”
In United States v. Weatherspoon, we held that the government improperly vouched for its witness where the prosecutor stated “three times over in rapid succession that a witness ‘told the truth.’ ” 410 F.3d 1142, 1148 (9th Cir. 2005) (noting that the “ethical bar is set higher for the prosecutor than for the criminal defense lawyer”). In this case, the prosecutor stated at least three times that Rosenberg told the truth or that his allegations were true ((1) “the truth in this case is that [Preston] put his penis in [Rosenberg’s mouth];” (2) “In the end ... the truth came out of the mouth of [Rosenberg]; and (3) “Rosenberg probably wishes that the [drugs] could erase this memory ... But they can’t ... Because this is the truth of what happened.”). As in Weatherspoon, this amounted to improper vouching. The district court abused its discretion in allowing the final comment over defense counsel’s objection. Accordingly, these statements will factor into our cumulative error analysis.
3. Misstatement of the Evidence
Finally, Preston argues that the government misstated the evidence in summation. A prosecuting attorney may not misstate or manipulate the evidence at trial. Darden v. Wainwright, 477 U.S. 168, 182, 106 S.Ct. 2464, 91 L.Ed.2d 144 (1986); United States v. Kojayan, 8 F.3d 1315, 1320-22 (9th Cir. 1993).
During summation and again in rebuttal, government counsel stated to the jury, without objection, that the defense was asking them to believe that Rosenberg is “making up these allegations because he is a vicious, cold, calculating human being,” and that “the only two possibilities in this case are that Rosenberg is telling the truth,” or that he “is truly despicable and evil and lying about everything for no reason.” This argument misstates the evidence. The defense presented no evidence suggesting that Rosenberg was a “vicious” or “evil” liar, nor was this line of attack part of any argument advanced by the defense. Instead, the defense simply provided expert opinion testimony that Rosenberg may have experienced memory problems as a consequence of his drug use. The defense’s memory expert, Dr. Simpson, testified that individuals, and especially drug addicts, can make “source monitoring errors,” whereby they confuse an imagined, dreamed, or hallucinated event with one that actually occurred. The prosecutor posed a false choice to the jury (that either what Rosenberg is saying is the truth, or else he is an evil liar). The government also misstated the nature of the evidence (which was actually that Rosenberg could be misremembering or could be telling the truth). It was plainly improper for the prosecutor to characterize the evidence in this manner, and the government’s statements will be factored intp our cumulative error analysis.
D. Cumulative Error Analysis
Having identified the trial errors in this case, we turn now to a discussion of whether their cumulative effect prejudiced Preston. Below is a summary of the errors at trial—including a description of any mitigating or aggravating factors (such as a curative instruction or a jury question indicating that a juror did not follow a curative instruction)—followed by an analysis of these errors within the context of the trial and in view of the strength of the government’s case.
1. Trial Errors
The first series of errors relates to the opinion testimony that bolstered the veracity of Rosenberg’s testimony. Bussart stated on three separate occasions that she believed Rosenberg’s allegations were true; Barry Rosenberg testified that he did not question his brother’s allegations because they made sense to him; and Agent Dellacroce testified that he did not believe Preston when he denied Rosenberg’s allegations. In addition, Bussart impermissibly offered her opinion that victims’ sexual abuse allegations are generally true and that Rosenberg’s emotions were similar to those of a person who has suffered sexual abuse.
The court did give a curative instruction following Bussart’s testimony, explaining to the jury that they must disregard any witness’s opinion about the truthfulness of another witness. However, after Agent Dellacroce testified that he did not believe Preston’s denial two separate jurors sought to ask Agent Dellacroce—who testified after Bussart—ivhy he did not believe Preston. This suggests that the jurors focused on Dellacroce’s opinion that Preston was not believable, and that they had trouble absorbing the court’s earlier instruction to disregard witness testimony about the truthfulness of others. See Avila v. Los Angeles Police Dept., 758 F.3d 1096, 1118 n.17 (9th Cir. 2014) (Vinson, J., dissenting) (jury questions can serve as evidence of the effectiveness of curative instructions).
Next, the trial court erred in allowing Ms. Preston to testify that in 2008 she saw Preston masturbating to a picture of his eight-year-old stepson in socks and underwear. In summation, the government reiterated her testimony and added that this evidence, “only goes to [Preston’s] intent and that’s it.”
Finally, the errors included the following conduct by the prosecutor: (1) commenting on Preston’s decision not to testify, (2) asserting at least three times that Rosenberg was telling the truth (which the prosecutor tried to mitigate after the fact, by reminding the jury that what attorneys argue is not evidence), and (3) suggesting inaccurately that the evidence presented a choice between either believing Rosenberg and finding Preston guilty, or necessarily finding that Rosenberg was a diabolical liar and that Preston was not guilty.
2. Cumulative Effect
The government correctly highlighted the key issue in this case in its summation to the jury by stating that, “[i]n this case really there’s only one question: do you believe Tim Rosenberg or not?” The cumulative effect of these trial errors prejudiced Preston because almost all of them implicated this very question. Because this cumulative effect is so clear, we do not decide the individual prejudice of any of the errors.
Three of the errors involve statements— by Bussart, Barry Rosenberg, and the prosecutor—either asserting or implying that Tim Rosenberg’s testimony was true. Two others, Agent Dellacroce’s testimony and the prosecutor’s comment on Preston’s decision not to testify, improperly suggested the inverse—that Preston’s denial of the allegations was false. Moreover, the prosecutor suggested that the jury could only find Preston not guilty by deciding that Tim Rosenberg—for whom the government and many of their witnesses had vouched—was a diabolical liar. In this way, the errors at trial were not isolated. Indeed, they stand in “unique symmetry such that they amplify each other in relation to [the] key,” and only, “contested issue in the case”—whether the alleged victim, Tim Rosenberg, was telling, the truth. Ybarra v. McDaniel, 656 F.3d 984, 1001 (9th Cir. 2011) (quoting Parle v. Runnels, 505 F.3d 922, 933 (9th Cir. 2007)).
The only error in this case that did not relate to bolstering Rosenberg as a truth-teller was hardly insignificant. Andrea Preston’s testimony about Preston’s masturbation suggested he had the character of a child molester. Where intent was not in dispute, allowing such inflammatory evidence likely had a prejudicial effect by encouraging the jury to believe that Preston was the type of person who would sexually abuse Rosenberg.
Moreover, the government’s case hinged almost entirely on Rosenberg’s testimony. The government points out that Preston’s denials of the allegations were not credible because he told Agent Dellacroce that he “could not remember” whether he received oral sex from Rosenberg. But Agent Della-croce equivocated on this point and the interview was not recorded. The only other witnesses present in .the Preston home when the molestation allegedly took place, Barry Rosenberg and Andrea Preston, testified that they never suspected anything. Thus, the only substantial proof that the government offered at trial was Rosenberg’s testimony.
In sum, there were multiple errors- in the trial that unfairly bolstered, the victim’s credibility, and an additional error, that portrayed- the defendant as -the “typ® of person” who would molest a child. In addition, the government’s case hinged entirely on Rosenberg’s credibility, with little additional proof to corroborate his allegations. The cumulative effect of these errors rendered Preston’s trial fundamentally unfair, and his conviction must therefore be reversed and the case remanded for a new trial. .
IV. Conclusion
For the foregoing reasons, the matter is' REVERSED AND REMANDED FOR PROCEEDINGS NOT INCONSISTENT WITH THIS OPINION.
. Based on the district court's preliminary ruling, a proper question for the prosecutor to ask Bussart would have been: “Is your opinion about Rosenberg's lack of truthfulness limited to his statements to you concerning his use of drugs and alcohol, or are you saying he was generally untruthful?”
. Where the other acts offered are specific incidents of prior child molestation—which would clearly be similar to the charged offense here—Federal Rule of Evidence 414 expressly permits them to be admitted “on any matter to which it is relevant.” Other acts of mere fantasy ‘or auto-eroticism, however, are not permitted under this rule.
. This, scenario stands in contrast to sexual misconduct cases where a defendant puts intent at issue and other act evidence of their deviant fantasies, therefore, increases in probative value. See, e.g., Curtin, 489 F.3d at 950 (child-sex stories in defendant’s possession probative of intent where he was charged with attempting to engage in a sexual act with a minor and admitted to talking to an individual online who identified as a minor but testified that he intended her to be an adult role-playing a teen); United States v. Cunningham, 607 Fed.Appx. 715 (9th Cir. 2015) (child pornography'that the defendant downloaded probative of intent where he was charged with intending to use a child to produce a sexually explicit visual depiction and admitted taking pictures of a child but testified that he did not intend for any of the pictures to be pornographic).
. Moreover, the prosecutor’s statement that there was no evidence contradicting Rosenberg’s testimony was, strictly speaking, incorrect: both Agent Dellacroce and Detective Rodriguez testified that, during their respective interviews of Preston, he denied Rosenberg's allegations.