The United States Supreme Court has never extended its holding in Herring v. New York, 422 U.S. 853, 95 S.Ct. 2550, 45 L.Ed.2d 593 (1975), to support the majority’s claim of structural error where the affirmative defense of duress necessarily requires admission of the criminal conduct sought to be excused. There is a fundamental difference between the complete denial of closing argument at issue in Herring and the limitations on closing argument imposed in Frost’s trial. By declaring structural error in this circumstance, our court once again ignores the Supreme Court’s trenchant instructions to exercise restraint in defining clearly established federal law, to grant deference to state courts under the Antiterrorism and Effective Death Penalty Act (“AEDPA”), and to find structural error only in rare instances. *920In light of these oft-repeated directives, we should conclude that the Washington Supreme Court’s interpretation of Herring, as inapplicable to reverse Frost’s convictions, was neither contrary to nor an unreasonable application of clearly established federal law, and that we are compelled under AEDPA to grant deference to the state court’s decision. See Harrington v. Richter, 562 U.S. 86, 131 S.Ct. 770, 785, 178 L.Ed.2d 624 (2011).
I
The majority’s interpretation of Hening is not the only reasonable reading of the Court’s opinion and, therefore, is not persuasive on AEDPA review. The majority relies on Herring to conclude that the trial court’s error — restricting Frost’s closing argument to permit either a failure-of-proof argument or a duress defense (and not both) — was structural and not subject to harmless-error analysis. That argument misconstrues and unjustifiably extends this Supreme Court precedent. Herring simply held that a court’s “total denial” of closing argument constituted structural error. Id. at 858-59, 863-65, 95 S.Ct. 2550. Herring is silent on whether an erroneous limitation requiring a defendant to choose between two incompatible defenses, such as the one imposed by the trial court in this case, is structural error.
In Herring, the defendant was denied any opportunity to make a closing argument before judgment was rendered in a criminal bench trial. Id. At the conclusion of the bench trial, Herring’s counsel asked “to be heard somewhat on the facts.” Id. at 856, 95 S.Ct. 2550. Relying on a New York statute that “confer[red] upon every judge in a nonjury criminal trial the power to deny counsel any opportunity to make a summation of the evidence before the rendition of judgment,” id. at 853, 95 S.Ct. 2550, the judge responded, “I choose not to hear summations,” id. at 856, 95 S.Ct. 2550. The judge then found Herring guilty of the crime charged. Id.
The Supreme Court vacated the conviction, holding that the utter denial of closing argument violated Herring’s right to counsel. Id. at 857-65, 95 S.Ct. 2550. The Court concluded that “there can be no justification for a statute that empowers a trial judge to deny absolutely the opportunity for any closing summation at all.” Id. at 863, 95 S.Ct. 2550 (emphasis added). The Supreme Court stated that “[tjhere can be no doubt that closing argument for the defense is a basic element of the adversary factfinding process in a criminal trial” and that “a total denial of the opportunity for final argument in a nonjury criminal trial” violates the Sixth Amendment. Id. at 858-59, 95 S.Ct. 2550. As a result, the Court held that denying counsel any opportunity to make a closing argument is structural error. Id.
In announcing its decision, however, the Court was careful to observe that there is a fundamental difference between a complete denial of closing argument and a limitation on the scope of closing argument. The Court acknowledged what every trial judge knows: that mere limitations on closing arguments do not constitute structural error because “[t]he presiding judge must be and is given great latitude in controlling the duration and limiting the scope of closing summations.” Id. at 862, 95 S.Ct. 2550. This is consistent with longstanding precedent under which we have held that a trial judge has broad discretion to control closing argument. See United States v. Guess, 745 F.2d 1286, 1288 (9th Cir.1984).
The complete denial of closing argument at issue in Herring cannot be equated with the limitations on closing argument imposed in Frost’s trial. Unlike in Herring, the trial judge allowed Frost’s counsel to make a closing argument. The judge re*921stricted the scope of that argument, incorrectly requiring counsel to choose between two conflicting defenses, but permitting counsel to argue one. Frost’s counsel elected to argue Frost’s duress. Herring does not compel the conclusion that a court’s ruling that restricts the scope of argument, but does not entirely prohibit closing argument, is structural error. Frost’s claim requires more than simply applying the rule announced in Herring to a different factual scenario. Application of Herring to Frost’s case inappropriately extends the Supreme Court’s holding to a novel legal question that it has never addressed.
II
A
The majority’s expansive interpretation of Herring cannot be reconciled with the Supreme Court’s instructions to only find structural error in rare instances. The Court has “repeatedly recognized that the commission of a constitutional error at trial alone does not entitle a defendant to automatic reversal. Instead, most constitutional errors can be harmless.” Washington v. Recuenco, 548 U.S. 212, 218, 126 S.Ct. 2546, 165 L.Ed.2d 466 (2006) (internal quotation marks omitted). “Only in rare cases has th[e] Court held that an error is structural, and thus requires automatic reversal.” Id. These “rare cases” involve the complete denial of counsel, a biased trial judge, a defective instruction defining proof beyond a reasonable doubt, denial of self-representation at trial, and racial discrimination in the selection of the grand jury. Id. at 218-19 n. 2, 126 S.Ct. 2546.
The Supreme Court has determined that “if the defendant had counsel and was tried by an impartial adjudicator, there is a strong presumption that any other [constitutional] errors that may have occurred are subject to harmless-error analysis.” Id. at 218, 126 S.Ct. 2546 (internal citations and quotations omitted). The Court has found the following errors to be subject to harmless-error review: jury instructions that misstate an element of the offense; improper comment on defendant’s silence at trial in violation of the Fifth Amendment Self-Incrimination Clause; failure to instruct the jury on the presumption of innocence; and failure to give a jury instruction on a lesser included offense in a capital case in violation of the Due Process Clause. Arizona v. Fulminante, 499 U.S. 279, 306-07, 111 S.Ct. 1246, 113 L.Ed.2d 302 (1991).
The Washington Supreme Court correctly concluded that “the error [committed by the trial court] [wa]s not so egregious as to require automatic reversal” and was comparable to the errors that the Supreme Court has previously found harmless. State v. Frost, 160 Wash.2d 765, 161 P.3d 361, 370 (2007) (en banc). The error “occurred during the presentation of the case to the jury, and [could] therefore be quantitatively assessed in the context of other evidence presented in order to determine whether [it] was harmless beyond a reasonable doubt.” Fulminante, 499 U.S. at 307-08, 111 S.Ct. 1246. Reviewing the record as a whole, the Washington Supreme Court correctly concluded that the error committed in Frost’s case was of similar magnitude and impact as other constitutional violations found by the United States Supreme Court to be harmless.
In the absence of a United States Supreme Court opinion holding that partial restrictions on closing argument amount to structural error, and in light of the Court’s precedent that most errors are subject to harmless-error review, the Washington Supreme Court’s conclusion that the error identified by Frost was harmless was not objectively unreasonable. In holding that *922it was, the majority not only fails to grant appropriate deference as required by AEDPA, but it also ignores the Supreme Court’s admonition to find structural error only in rare and limited circumstances. When it elected to extrapolate a novel rule from Herring that was neither considered nor addressed by the Supreme Court, the majority failed to heed the Supreme Court’s warning that constitutional errors are strongly presumed to be harmless.
B
The majority concludes that we must infer a broader rule from Herring — that structural error also occurs when a trial court limits the scope of closing argument, prohibiting a defendant from arguing a defense or claim but permitting closing argument on the primary defense theory. But the Supreme Court has repeatedly admonished us not to infer extensions from the rules identified in its opinions. See, e.g., Nevada v. Jackson, — U.S. -, 133 S.Ct. 1990, 1994, 186 L.Ed.2d 62 (2013) (per curiam); Richter, 131 S.Ct. at 785.
The fact that a rule might be necessarily implied from a Supreme Court decision is insufficient to show that the rule is clearly established federal law under AEDPA. See, e.g., Wright v. Van Patten, 552 U.S. 120, 124-26, 128 S.Ct. 743, 169 L.Ed.2d 583 (2008) (per curiam) (holding that no prior Supreme Court decision provided a “categorical answer” to the question of whether prejudice may be presumed when defense counsel participated in a plea hearing by telephone, stating “[b]ecause our cases give no clear answer to the question presented ... it cannot be said that the state court unreasonably] appli[ed] clearly established Federal law”); Kane v. Garcia Espitia, 546 U.S. 9, 10, 126 S.Ct. 407, 163 L.Ed.2d 10 (2005) (per curiam) (“[I]t is clear that [a defendant’s right to self-representation under] Faretta does not, as § 2254(d)(1) requires, ‘clearly establis[h]’ the law library access right[;] [i]n fact, Faretta says nothing about any specific legal aid that the State owes a pro se criminal defendant.”).
In Jackson, the Supreme Court unanimously reversed us for extending the rules announced in its opinions. 133 S.Ct. at 1990-94. The three judge Jackson panel had relied upon “Supreme Court decisions holding that various restrictions on a defendant’s ability to cross-examine witnesses violate the Confrontation Clause of the Sixth Amendment.” Id. at 1994 (emphasis in original). The panel then inferred from these cases the specific rule that “the Confrontation Clause entitles a criminal defendant to introduce extrinsic evidence for impeachment purposes.” Id. (emphasis in original).
Our Jackson panel made this leap in logic, “elid[ing] the distinction between cross-examination and extrinsic evidence by characterizing the cases as recognizing a broad right to present ‘evidence bearing on [a witness’] credibility.’” Id. (second alteration in original). In reversing that opinion, the Supreme Court pointedly stated that “[b]y framing our precedents at such a high level of generality, a lower federal court could transform even the most imaginative extension of existing case law into ‘clearly established Federal law, as determined by the Supreme Court.’ ” Id.
The majority repeats the same mistake committed by our Jackson panel. It has extended the narrow holding in Herring to find structural error where a defendant is prevented from arguing in closing one of several inconsistent defense theories. It does this despite the fact that the Court has repeatedly and expressly instructed us to exercise restraint when defining clearly established federal law — an instruction that the majority all too readily ignores. See, e.g., Richter, 131 S.Ct. at 785; see also Williams v. Taylor, 529 U.S. 362, 385-86, *923120 S.Ct. 1495, 146 L.Ed.2d 389 (2000) (stating that AEDPA created a “ ‘mood’ that the Federal Judiciary must respect” in that “federal judges [should] attend with the utmost care to state-court decisions ... before concluding that those proceedings were infected by constitutional error sufficiently serious to warrant the issuance of the writ”).
AEDPA deference requires that a claim be based upon a Supreme Court decision that existed at the time of the state court adjudication. The Washington Supreme Court correctly concluded that the United States Supreme Court’s decision in Herring did not extend to restrictions on closing argument that fall short of outright prohibitions. We should abide by the AEDPA statutory restriction and accord Washington’s highest court the comity that Congress and the Supreme Court require.
C
The logical extension of the majority’s rule would be to declare structural error and require automatic reversal any time a trial judge erred in placing limits on closing argument because petitioner could argue that, as to the contested issue, the limitation resulted in a total denial of closing argument on a legitimate defense theory. This result is apparent when you look below the surface of the majority’s all-or-nothing argument, which amounts to stating, “If you don’t like your closing argument and can find any error, no matter how small, we won’t hold you to your conviction.” This logic would require convictions to be vacated and new trials granted on a number of lesser errors, which may now be deemed structural.
For example, if structural error is found in this case, how could we say that structural error does not also occur when a trial judge improperly excludes exculpatory evidence without regard to materiality and forbids defense counsel from commenting on it in closing? In both instances, the defendant is prohibited from effectively claiming innocence and from comprehensively arguing all theories that support that defense. Structural error may also be found when a trial judge improperly restricts closing argument by imposing time limitations that necessarily require the defense to choose among multiple arguments to emphasize during summation. Claims of structural error may also be made when a judge adopts any number of limitations that inhibit defense counsel from making their most effective and comprehensive arguments supporting a defendant’s innocence without regard to any evidence supporting the claim or the context from which the argument is to be made. That cannot be the law and the Supreme Court has never said anything of the sort.
There are substantial negative consequences to interpreting the Supreme Court’s decision in Herring expansively and increasing the number of errors deemed structural. Doing so eliminates any need to establish prejudice to show that the defendant was deprived of a fair trial. The majority’s interpretation, and its undeniable effect, cannot be reconciled with our limited role in habeas proceedings involving review of state convictions under AEDPA.
Ill
The majority cites to two Ninth Circuit cases, Conde v. Henry, 198 F.3d 784 (9th Cir.1999), and United States v. Miguel, 338 F.3d 995 (9th Cir.2003), to conclude “that preventing a defendant from arguing a legitimate defense theory constitutes structural error.” The majority’s reliance on these cases is in error. As the district court held, “Circuit case law is not clearly established federal law as determined by the Supreme Court and is not, alone, a *924basis for the Court to grant habeas relief.” See Renico v. Lett, 559 U.S. 766, 779, 130 S.Ct. 1855, 176 L.Ed.2d 678 (2010) (stating that a decision of the court of appeals “does not constitute clearly established Federal law, as determined by the Supreme Court, § 2254(d)(1), so any failure to apply that decision cannot independently authorize habeas relief under AEDPA.” (internal quotation marks omitted)); see also Parker v. Matthews, — U.S. -, 132 S.Ct. 2148, 2155, 183 L.Ed.2d 32 (2012) (per curiam) (stating that “circuit precedent does not constitute clearly established Federal law,” and it “cannot form the basis for habeas relief under AEDPA”). Consequently, our decisions in Conde and Miguel are insufficient to find that the Washington Supreme Court’s decision — that the trial court’s error was not structural — was contrary to, or an unreasonable application of, clearly established federal law.
Furthermore, even if Conde or Miguel were instructive in defining clearly established federal law, both cases are distinguishable. Importantly, neither case involved a petitioner seeking habeas relief post-AEDPA, so neither case bound our court to AEDPA’s deferential review standards on appeal. On federal habeas review, we must uphold the Washington Supreme Court’s adjudication of Frost’s claim unless we conclude that its interpretation limiting the extension of Herring was objectively unreasonable, and not merely because we would have reached a contrary interpretation. See Renico, 559 U.S. at 776, 130 S.Ct. 1855 (“AEDPA authorizes federal courts to grant relief only when state courts act unreasonably.”).
In Conde, a pre-AEDPA habeas appeal, we concluded that structural error occurred because “the trial court improperly precluded Conde’s attorney from making closing argument explaining the defendant’s theory of the case, it refused to instruct the jury on the defendant’s theory, and, over the defendant’s objection, it gave jury instructions that did not require that the jury find every element of the offense.” 198 F.3d at 741. “Together,” we found, “these errors deprived the petitioner of effective assistance of counsel, due process and trial by jury on every element of the charged crime.” Id.
In Miguel, a direct appeal from a federal criminal proceeding, we held that structural error occurred when the district court precluded defense counsel from arguing during closing that someone other than the defendant shot the victim and that no evidence supported the defense theory. 338 F.3d at 1000-01. Although we cited to Herring in so ruling, our decision in Miguel insufficiently establishes under AEDPA that the Washington Supreme Court’s determination in this case was contrary to, or an unreasonable application of, clearly established federal law.
Here, after he had taken the stand and admitted to participating in an eleven-day crime spree, Frost was permitted to argue during closing argument that he committed the crimes under duress — the primary defense theory in his case. Further, as the Washington Supreme Court and the district court highlighted, “[t]he record clearly shows the prosecutor argued it was the state’s burden to prove that Frost was an accomplice and to prove beyond a reasonable doubt each and every element of the charged offenses.” Frost v. Van Boening, No. C09-725-TSZ-BAT, 2010 WL 5775657, at *8 (W.D.Wash. Oct. 5, 2010) (adopted by Frost v. Van Boening, No. C09-725Z, 2011 WL 486198 (W.D.Wash. Feb. 4, 2011)); see Frost, 161 P.3d at 364 (“In closing, the prosecutor repeatedly mentioned the State’s burden of proof as to Frost’s robbery offenses. Likewise, the jury was properly instructed on the State’s burden of proof in general, as well as the requirements to prove accomplice liability *925in particular.” (internal citation omitted)). Thus, the jury received ample instruction on the cardinal principles of criminal law and it was properly equipped to apply the facts, as it found them, to the legal framework.
By relying on circuit precedent to find structural error, the majority also ignores the Supreme Court’s recent admonition in Marshall v. Rodgers, — U.S. -, 133 S.Ct. 1446, 185 L.Ed.2d 540 (2013) (per curiam). In reversing us in Marshall, the Court noted that our opinion “rested in part on the mistaken belief that circuit precedent may be used to refine or sharpen a general principle of Supreme Court jurisprudence into a specific legal rule that this Court has not announced.” Id. at 1450. It reiterated that in reviewing habe-as petitions we “may not canvass circuit decisions to determine whether a particular rule of law is so widely accepted among the Federal Circuits that it would, if presented to this Court, be accepted as correct.” Id. at 1451.
This is precisely the tactic that the majority has employed to grant Frost’s habe-as petition. Given our limited review under AEDPA, our decisions in Conde and Miguel do not establish that the Washington Supreme Court’s decision to apply harmless-error review was contrary to or involved an unreasonable application of clearly established federal law, as determined by the Supreme Court. Absent such a showing, under AEDPA we must defer to the state court’s decision.
IV
The majority also bends the facts of the case in its favor by framing a choice as a mandate. The majority asserts that the judge “specifically prohibited counsel from arguing that the State had not met its burden of proof’ and that “defense counsel was precluded from arguing reasonable doubt.” But that is incorrect. The trial judge repeatedly indicated that Frost was free to assert that the prosecution had not proved accomplice liability. But the judge also noted, albeit in error, that such a choice would have consequences, informing Frost that “[i]f he [argued failure of proof during closing,] the duress instruction will come out of the case.” By requiring Frost to “opt for one [theory] or the other,” the judge presented a choice to Frost as opposed to handing down a prohibition or mandate. And, although the judge may have hinted at his belief that abandoning the duress defense and electing to argue that the prosecution had not satisfied its burden of proof would be unwise, the judge did not usurp Frost’s choice.1
This is material to the analysis because the majority finds support for its structural-error conclusion based on Frost’s actions instead of the judge’s. The majority concludes that by “depriv[ing] Frost of his right to ‘insist that his guilt be established *926beyond a reasonable doubt,’ ” the trial judge essentially entered a “directed verdict on guilt.” Relying on Supreme Court cases that instruct both (1) that a defendant cannot constitutionally be tried under a lesser burden of proof and (2) that due process prevents a state from shifting the burden of proof to the defendant, the majority concludes that “the trial judge’s actions took the question of reasonable doubt away from the jury” such that the “case was impermissibly and unconstitutionally tried to the jury with the burden of proof on Frost.” But this overlooks Frost’s role. After all, had Frost unwisely pursued his failure-of-proof argument in lieu of his duress defense, these ancillary constitutional issues that the majority leans on would be deprived of their supporting role.
V
AEDPA mandates that “a state prisoner seeking a writ of habeas corpus from a federal court ‘must show that the state court’s ruling on the claim being presented in federal court was so lacking in justification that there was an error well understood and comprehended in existing law beyond any possibility for fairminded disagreement.’ ” Bobby v. Dixon, — U.S. -, 132 S.Ct. 26, 27, 181 L.Ed.2d 328 (2011) (per curiam) (quoting Richter, 131 S.Ct. at 786-87). In the absence of a Supreme Court holding that the error in this case was structural error — and in light of the Court’s recurring precedent that most errors are subject to harmless-error analysis and that clearly established federal law must be interpreted narrowly — the Washington Supreme Court’s conclusion that harmless error applies here was not objectively unreasonable.
AEDPA does not permit us to reject a state court’s interpretation of Supreme Court precedent simply because we disagree. See, e.g., Lockyer v. Andrade, 538 U.S. 63, 75, 123 S.Ct. 1166, 155 L.Ed.2d 144 (2003). In Lockyer, the Supreme Court reversed our decision to grant habe-as relief, explaining:
It is not enough that a federal habeas court, in its “independent review of the legal question,” is left with a “firm conviction” that the state court was “erroneous.” We have held precisely the opposite: “Under § 2254(d)(l)’s ‘unreasonable application’ clause, then, a federal habeas court may not issue the writ simply because that court concludes in its independent judgment that the relevant state-court decision applied clearly established federal law erroneously or incorrectly.” Williams v. Taylor, 529 U.S. 362, 411 []120 S.Ct. 1495, 146 L.Ed.2d 389 (2000). Rather, that application must be objectively unreasonable. Id. at 409 [120 S.Ct. 1495]; Bell v. Cone, 535 U.S. 685, 699 [122 S.Ct. 1843, 152 L.Ed.2d 914] (2002); Woodford v. Visciotti, 537 U.S. 19, 27 [123 S.Ct. 357, 154 L.Ed.2d 279] (2002) (per curiam).
Id. at 75-76, 123 S.Ct. 1166 (selected internal citations and quotation marks omitted); see also Schriro v. Landrigan, 550 U.S. 465, 473, 127 S.Ct. 1933, 167 L.Ed.2d 836 (2007) (“The question under AEDPA is not whether a federal court believes the state court’s determination was incorrect but whether that determination was unreasonable — a substantially higher threshold.”).
Because the Supreme Court has never addressed a claim, such as the one presently before us, concerning a restriction on the scope of closing argument, the Washington Supreme Court’s determination that the error was not structural does not require automatic reversal. See Richter, 131 S.Ct. at 786 (“[I]t is not an unreasonable application of clearly established Federal law for a state court to decline to apply a specific legal rule that has not been squarely established by this Court.” (quoting Knowles v. Mirzayance, 556 U.S. *927111, 122, 129 S.Ct. 1411, 173 L.Ed.2d 251 (2009) (alteration in original))).
The Supreme Court’s decision in Herring did not “establish a legal principle that dearly extends” to limitations imposed on closing arguments, as opposed to the complete denial of argument. Moses v. Payne, 555 F.3d 742, 754 (9th Cir.2009) (emphasis added). Indeed, the Washington Supreme Court was able to “draw a principled distinction between the case before it and Supreme Court case law,” and we are bound under AEDPA to defer to the state court’s reasoned opinion. Murdoch v. Castro, 609 F.3d 983, 991 (9th Cir.2010) (en banc);. see also Van Patten, 552 U.S. at 124-26, 128 S.Ct. 743.
VI
The Washington Supreme Court properly concluded that the trial court’s error was harmless. On habeas review, to determine whether Frost is entitled to relief, our analysis is focused on whether the trial court’s restriction on Frost’s closing argument “had [a] substantial and injurious effect or influence in determining the jury’s verdict.” Brecht v. Abrahamson, 507 U.S. 619, 637, 113 S.Ct. 1710, 123 L.Ed.2d 353 (1993) (quoting Kotteakos v. United States, 328 U.S. 750, 776, 66 S.Ct. 1239, 90 L.Ed. 1557 (1946)).
Under this standard, Frost is “not entitled to habeas relief based on trial error unless [he] can establish that it resulted in ‘actual prejudice.’ ” Id. (quoting United States v. Lane, 474 U.S. 438, 449, 106 S.Ct. 725, 88 L.Ed.2d 814 (1986)). We should hold that, “[i]n light of the record as a whole,” the trial court’s limitation on defense counsel’s closing argument did not have a “substantial and injurious effect or influence in determining the jury’s verdict” for several reasons. Id. at 638, 113 S.Ct. 1710 (internal quotation marks omitted). That is what the Washington Supreme Court ultimately, and correctly, decided. See Frost, 161 P.3d at 364.
First, the evidence of Frost’s guilt at trial was overwhelming. Frost gave three taped confessions, all of which were entered into evidence at trial. Further, Frost testified in detail about his involvement in the crimes for which he was charged as an accomplice, admitting that: (1) he drove Williams to the Gapp residence on the night of the robbery, that he entered the residence, and removed money and guns from Gapp’s safe; (2) he drove the codefendants to the Taco Time restaurant, T and A video store, the 7-Eleven convenience store, and Ronnie’s Market on the night that each of those robberies took place; (3) he was aware that Williams carried a bag containing such items as a ski mask and gloves that were routinely used to commit the crimes; and (4) he was aware of his co-defendant’s use of firearms. Finally, Detective Broggi’s testimony regarding the loaded guns, cash register, bank bags, safe, and ski masks that were found at Frost’s home further corroborated Frost’s active role in the crimes.
Second, Frost conceded guilt as to some of these crimes when he testified in his own defense before the trial court erroneously restricted Frost’s closing argument. During opening statements, Frost’s counsel admitted that Frost committed at least one of the robberies, stating, “You will find from the evidence ... that Joshua Frost is guilty of the robbery of [the elderly couple].” Consistent with his counsel’s statements, Frost told the jury that he participated in the charged crimes but claimed that he did so under duress.
Third, as to the remaining offenses, the state’s burden of proof did not go untested because defense counsel was barred from raising reasonable doubt as to accomplice liability. Although Frost’s counsel admitted during closing argument that Frost *928was guilty of accomplice liability on certain counts, he also argued that the state had failed to meet its burden of proof on others. Specifically, Frost’s counsel argued:
I think you can find Joshua Frost guilty of the Gapp robbery because that is just so overpowering, and he did go into the house. I think you can find Joshua Frost guilty of the T and A robbery not because he went in to do the robbery but because he actually entered the store. And the only reason I think you could find him guilty of that is that it is kind of just too much to ask for somebody who is willing to take the step to let him off. And I know that is what you are thinking, some of you. But as to the cases in which he didn’t go in anywhere and was just told to stay put, we are asking you to find him not guilty, and even if you find him guilty he is not guilty of the guns. You can find him guilty of displaying the gun as an accomplice, I suppose, which is one of the things you have to find to make a robbery in the first degree. But that doesn’t require you to find the special verdict firearm allegation in addition. You don’t have to do that. And we hope you don’t. And we think that the basis for not doing that is that the guns were out of his control.
Accordingly, as the district court acknowledged, “contrary to Frost’s contention, the record shows defense counsel was able to and did argue the state had failed to prove Frost was an accomplice.”
Fourth, the jury was fully informed of the state’s burden to prove each element of the crime beyond a reasonable doubt, including accomplice liability. For example, the prosecutor in closing argument stated:
Now I have divided my closing argument into two different parts. The reason for that, ladies and gentlemen, is there really are two parts in some ways you look at this. The first part has to do with the charges and the evidence and has the state proven all of the elements and all of the crimes beyond a reasonable doubt. And then the second part has to do with the defense of duress, and this is important because the first part, again, the state has the burden. To get to duress you really have to find the state proves its case beyond a reasonable doubt....
Additionally, the jury in Frost’s trial was properly instructed as to the state’s burden of proof by the judge prior to opening statements, by the prosecution in closing, and in the formal jury instructions read before the jury rendered its verdict.
Finally, Frost voluntarily elected to concede his involvement in the robberies when he decided to pursue the inconsistent defense of duress. Under Washington law, a defendant must admit that he “participated in the crime” in order to argue duress. Wash. Rev.Code § 9A.16.060(l)(a). As the Washington Supreme Court explained, “a defense of duress admits that the defendant committed the unlawful act, but pleads an excuse for doing so ... [and] a duress defense necessarily allows for no doubt that the defendant did the acts charged.” State v. Riker, 123 Wash.2d 351, 869 P.2d 43, 52 (1994) (emphasis in original).
Therefore, although the Washington Supreme Court found that the trial court erred when it required Frost to concede guilt or criminal liability, by electing to raise a duress argument, Frost necessarily had to admit that he committed the unlawful acts, and he consequently knew that any argument he made regarding the state’s failure to prove guilt was significantly weakened. Although the trial judge indisputably erred in prohibiting defense counsel from arguing innocence, the magnitude of that error is certainly lessened by the fact that Frost voluntarily admitted *929his involvement in each of the charged crimes by acknowledging his commission of them under oath while pursuing a duress defense.
Thus, in light of our review of the record, we should hold that the trial court’s limitation on defense counsel’s closing argument did not have a “substantial and injurious effect or influence in determining the jury’s verdict.” Brecht, 507 U.S. at 637, 113 S.Ct. 1710. There may be circumstances where a trial court’s limitation on closing arguments may not survive the Brecht harmless-error analysis. That is not the case here. Consequently, because Frost did not establish “actual prejudice,” we should conclude that he is not entitled to habeas relief. Id.
VII
The Washington Supreme Court’s decision that the trial court’s restriction on closing argument did not constitute structural error was neither contrary to, nor an unreasonable application of, clearly established federal law as determined by the United States Supreme Court. In light of the Supreme Court’s express instructions to exercise restraint in defining clearly established federal law, to grant deference to state courts under AEDPA, and to find structural error only in rare instances, the Washington Supreme Court’s decision to apply harmless-error analysis should have been respected.
The Washington Supreme Court properly concluded that the trial court’s error was harmless. Reviewing the record as a whole, the trial court’s restriction on Frost’s closing argument did not have a “substantial and injurious effect or influence in determining the jury’s verdict.” Brecht, 507 U.S. at 637, 113 S.Ct. 1710. The record demonstrates that Frost was afforded the opportunity to present his primary defense — duress—and the state was not relieved of its burden to prove Frost guilty beyond a reasonable doubt. Because the Washington Supreme Court’s decision is neither contrary to nor an unreasonable application of Supreme Court law as set forth in Herring, Frost is not entitled to habeas relief.
I respectfully dissent.
. The majority asserts that any choice Frost had “was a Hobson’s choice, the result of which allowed the burden of proof to be shifted to Frost.” But a Hobson’s choice, which requires selection between something or nothing, is only present here if we view Frost as being forced to choose between pursuing an affirmative duress defense or mounting no defense at all. This may be true given that Frost had already thrice confessed and then testified to participating in the crimes, but if so, any Hobson’s choice was of his own making. As opposed to a Hobson’s choice, Frost’s predicament is more akin to "Morton’s Fork.” Burroughs v. Metro Goldwyn-Mayer, Inc., 683 F.2d 610, 623 n. 13 (2d Cir.1982) (describing "Morton’s Fork” in the context of a different legal dispute). The trial court erroneously made Frost choose between two theoretically opposite times, both of which led to an equally undesirable outcome — conviction. Upon cry of "error,” and at the public’s expense, the majority suspends reality and charts Frost a new course around both Scylla and Charybdis. AEDPA prevents such safe passage when the resulting shipwreck was inevitable.