On December 18, 1974, another panel of this court (Circuit Judges Wright and Goodwin and District Judge Schwartz) reversed a denial of this same defendant’s motion for a new trial.
The government thereafter represented to this court, on a petition for rehearing, that an unresolved conflict in the evidence supporting the underlying motion had been discovered and that this conflict required an evidentiary hearing. The court withdrew its opinion and remanded the cause for the evidentiary hearing. That hearing was concluded in due course, and the district court made findings and conclusions. Once again the district court denied the motion for a new trial. That denial is challenged in this appeal.
Because the essential facts and the correct statement of the applicable law appear in the December 18, 1974, opinion, we adopt it as our own and incorporate the text substantially as originally filed **:
WRIGHT, Circuit Judge:These appellants are again before us, this time appealing from an order denying their motions for a new trial under Rule 33, Rules of Criminal Procedure. Their earlier convictions had been affirmed on appeal. United States v. Butler, 472 F.2d 1 (9th Cir.), cert. den. 414 U.S. 864, [94 S.Ct. 37, 38 L.Ed.2d 84], reh. den. 414 U.S. 1033, [94 S.Ct. 463, 38 L.Ed.2d 325] (1973).
The thrust of appellants’ contentions is that, after their earlier appeal, they learned of government promises to its key witness, John Durden, that pending charges against him would be dismissed if he testified favorably to the prosecution and against the Butlers. The district court conducted a hearing on the new trial motion, concluded that the new discovered evidence would not have altered the jury’s verdict, and denied the motion. We reverse.
*887In our earlier opinion, we noted that: The principal witness for the government was John Durden who had been caught by federal narcotics agents and apparently decided to cooperate by assisting the agents in their efforts to secure evidence against his fellow conspirators in hopes of receiving lenient treatment for himself. During the course of the trial, Durden’s credibility was frequently and vigorously attacked. One of the principal lines of assault comprised asking Durden whether in fact the pending indictment against him was going to be dismissed in consideration of his testimony against the defendant. To this question, he answered ‘no.’
Appellants- argue that, by allowing such testimony to go uncorrected, the prosecutor was acquiescing in the knowing use of false evidence in the manner forbidden by Napue v. Illinois, 360 U.S. 264, 79 S.Ct. 1173, 3 L.Ed.2d 1217 (1959), and more recently condemned in Giglio v. United States, 405 U.S. 150, 92 S.Ct. 763, 31 L.Ed.2d 104 (1972). However, such an argument is based upon the premise that a promise of dismissal had definitely been made to Durden and that this alleged promise was known to the prosecution. The government has at all times denied such a promise, and there is no evidence in the record to the contrary.
United States v. King, 472 F.2d 1, 5-6 (1973).
At the hearing on their new trial motion appellants were able to produce substantial new evidence which strongly supported their earlier charges. It consisted of affidavits and testimony of four defense attorneys associated with the Butler and Durden cases, including Saul Bernard, Durden’s attorney. There were also contemporaneous government file memoranda and the tape of surreptitiously recorded conversations between Dur-den and appellant Robert Butler.
This evidence raised a serious question as to the veracity of Durden’s testimony at the original trial that he had been given no promises of assistance in return for his testimony. It also cast doubt upon corroborative testimony and assurances, by the prosecutor and two government agents assigned to the case, to the effect that Durden had been promised nothing except that his cooperation would be brought to the attention of the judge in subsequent proceedings against him.1
The gist of the testimony by Saul Bernard was that agents Clemente and Sternaman had indicated on several occasions that dismissal or at least reduction of the charges pending against Durden was a strong probability. The court agreed with this assessment.2 Moreover,
the court specifically rejected the testimony of the assistant United States attorney who tried the Butler case. He testified that he had never discussed dismissal of Durden’s case with others on the prosecutorial staff, nor suggested to defense counsel-that dismissal was likely.3
*888Bernard’s affidavit is significant. His sworn statement that the assistant United States attorney had assured him that Dur-den’s federal case would be dismissed sharply contradicts the prosecuting attorney’s denial that he even talked with Bernard, much less discussed dismissal with him. The court concluded that a conversation between Bernard and the prosecutor had taken place and that dismissal was discussed. It further concluded that this discussion may have been such as to signal the anticipated dismissal to defense counsel.4
Bernard’s affidavit also maintained that the government agents had promised dismissal if Durden cooperated. That this promise was communicated to Durden is evident from Durden’s admissions to Butler, secretly taped by the latter, that the agents had given him substantial assurance of benefits greater than those they admitted at trial. While Durden’s recorded statement did not claim an actual promise existed, he did assert that the agents made it clear by innuendo that he could expect very lenient treatment following his testimony.
The lower court declined to attach much credence to the recording. Nevertheless, it found:
The Government agents had told Dur-den — at least that’s their statement— they told Durden he would help himself by cooperating with the Government and that they would call this cooperation to the attention of the Judge. Now, I think the agents told him more than that. The agents probably told Durden about the possible action that the court might take or could take and that they would make every effort to help him and I think that they did make every effort to help him because by their own testimony after the trial they were the ones that urged dismissal.
Durden said to Robert Butler that the agents said that they couldn’t make any definite promises, but not to worry, everything would be all right, and there were several comments Dur-den made along those lines.
I don’t doubt but that the agents made promises of benefits, without saying definitely — they had no authority to make any definite promise, that they were going to do everything they could to help him and I think they told him that.
These findings are consistent with Durden’s admissions and emphasize the district court’s conviction that both Dur-den and the government agents suppressed in their trial testimony the real nature of their dealings.
We now address the question whether this apparent effort of the prosecution to conceal the true nature of these dealings with its key witness should have been found adequate to support a new trial order.
Our opinion in United States v. Gerard, 491 F.2d 1300 (9th Cir. 1974), is controlling. There, we set out the “basic principle” governing this type of situation:
[T]he government is obliged to disclose pertinent material evidence favorable to the defense, and this applies not only to matters of substance, but to matters relating to the credibility of government witnesses. Giglio v. United States, 405 U.S. 150, [92 S.Ct. 763, 31 L.Ed.2d 104] (1972).
491 F.2d at 1302.
We drew a distinction in Gerard between cases where the evidence was suppressed inadvertently or otherwise in good faith, and those in which “improper prosecutorial motives” played a part in the suppression. In the former situation, we found that “if the suppressed evi*889dence was of unquestionable materiality, the government is not saved by good motives,” 491 F.2d at 1302, citing Brady v. Maryland, 373 U.S. 83, 87, [83 S.Ct. 1194, 10 L.Ed.2d 215] (1963). “On the other hand, when the suppressed evidence is of less obvious materiality, improper prosecutorial motives may be relevant,” either as an admission of materiality, or because we might choose to reverse “as a prophylactic against willful prosecutorial misconduct.” 491 F.2d at 1302, 1303.
The district court clearly found that both the prosecutor and the government agents gave more assurances to Durden and his attorney than they or Durden had been willing to adirtit at trial. These involved more than the usual pledge that Durden’s cooperation would be brought to the attention of the court. The prosecutor and the agents promised substantial benefits reasonably calculated, to be understood as probable dismissal or, failing that, certain misdemeanor disposition. This distinction might well be of importance to a jury charged, among other things, with weighing the credibility of the principal prosecution witness.
Moreover, even if the prosecutor’s conduct could be explained by a lack of knowledge of promises made to his principal witness, he would still be responsible for the consequences of his nondisclosure. The Supreme Court said in Giglio v. United States, 405 U.S. 150, 154, [92 S.Ct. 763, 31 L.Ed.2d 104] (1972):
The prosecutor’s office is an entity and as such it is the spokesman for the Government. A promise made by one attorney must be attributed, for these purposes, to the Government. .
To the extent this places a burden on the large prosecution offices, procedures and regulations can be established to carry that burden and to insure communication of all relevant information on each case to every lawyer who deals with it.
Consequently, this is a proper case for ordering a new trial on the basis of pros-ecutorial misconduct. Prosecution’s acquiescence in and corroboration of Dur-den’s denials at trial were followed by continued deception and half-truths on appeal and at the hearing on appellants’ new trial motion.
In short, we think the trial judge’s findings establish prosecutorial misconduct of the sort that both sides seem to concede would justify reversal. However, since the district judge denied the new trial motion on the basis that the newly discovered evidence would not have altered the jury’s verdict, we find an alternate holding proper in this case. Hence we hold that even if the court’s findings were found only to establish negligent or unintentional instead of willful nondisclosure of the prosecution’s dealings with Durden, reversal of the order denying a new trial would still be required.
It is conceivable that the judge might have found that, despite the evasive nature of the testimony of the prosecutor and his staff, they may have reasonably felt they were in literal compliance with proper standards. For example, even if Durden had been led to understand he would get a dismissal in return for his testimony, the court might have found no actual promise had been made.
Of course, if we felt the findings regarding prosecutorial misconduct were ambiguous, we could remand for more specific findings. That is unnecessary here.
Reversal follows whether the findings are interpreted as establishing bad faith or simply misinterpretation of the defense request, because the knowledge of the undisclosed dealings with Durden might reasonably have affected the jury’s assessment of his credibility.
This test results from the language in Giglio v. United States, quoting an earlier case to hold that:
A new trial is required if “the false testimony could ... in any reasonable likelihood have affected the judgment of the jury. . .
405 U.S. at 154, [92 S.Ct. 763], citing Napue v. Illinois, 360 U.S. 264, 271, [79 S.Ct. 1173, 3 L.Ed.2d 217] (1959).
Under the interpretation of this test set out in the district judge’s order, a *890new trial would be required only if the nondisclosure could reasonably have affected the judgment of the jury to the point of changing its verdict.
This reading of the Giglio language, in effect collapsing the new trial standard into the ordinary one for granting new trials, is the less attractive one. The thrust of Brady and Giglio, as well as of the proposed amendments to the Federal Rules of Criminal Procedure, 62 F.R.D. 271 (1974), seems to be in the direction of requiring more complete disclosure of evidence essential for, or at least useful to, full presentation of the defense.
The proper standard in negligent nondisclosure cases should call for a new trial wherever the nondisclosed evidence might reasonably have affected the jury’s judgment on some material point, without necessarily requiring a supplementary finding that it also would have changed its verdict. Giglio’s “effect on the judgment of the jury” language seems to focus on the quality of the evidence rather than on the outcome of the case.
This reading is supported by our holding in Flores v. Craven, 503 F.2d 1030 (9th Cir. Sept. 10, 1974), (memorandum for publication), where this court stated that the test under Giglio must be whether the nondisclosure “could have affected the [trier of fact’s] estimation of credibility.” See also Armour v. Salisbury, 492 F.2d 1032, 1036-37 (6th Cir. 1974); United States v. Diaz-Rodriguez, 478 F.2d 1005, 1008 (9th Cir. 1973) [test for granting new trial whether admission of undisclosed evidence would have been more likely to result in a different verdict].
Applying this test to the matter before us, we see that Durden’s testimony and credibility were crucial to the government’s case against appellants. Though certain “corroborative” tapes and ledgers were in evidence, it was the Durden testimony which made them legally and factually relevant and established the atmosphere in which they were interpreted.
We have thus far confined our discussion to the effect on the jury’s assessment of Durden’s credibility due to the tendency of the nondisclosure to enhance unduly the case against appellants. In addition to this, the suppression of material evidence served also to prejudice presentation of the defense case to the jury. The prosecution’s concealment of its dealings with Durden induced the trial judge to limit the scope of what could be argued.
We addressed this corollary contention in our earlier opinion, concluding that the trial judge did not commit reversible error in refusing to allow defense counsel to impeach Durden by arguing to the jury that Durden had been promised a dismissal in return for his testimony. We held that under the circumstances then known to us, the trial court’s ruling that defense counsel was entitled only to argue that Durden had as “great expectations of a dismissal as anyone in a similar position would have” was not unduly restrictive. 472 F.2d at 6.
We have held that the finding that prosecution countenanced false testimony in the presentation of its case of itself constituted sufficient grounds for granting a new trial. We also conclude that the motion for a new trial should have been granted because of the prejudicial effect of the nondisclosure of the requested information on the defense’s ability to present its case fairly. While argument and speculation by counsel cannot be considered as evidence, the defense was denied the opportunity to present impeaching evidence which could seriously have affected the jury’s assessment of the key prosecution witness’ credibility.
The government points out that the prosecution promise that it would bring Durden’s cooperation to the attention of the judge was adduced at trial. Any new evidence brought out at the motion for new trial would, it asserts, be merely cumulative to the argument defense counsel was allowed to make.
Applying the standard we hold applicable, we must reject this contention. Both by casting doubt on the prosecution case and by increasing the scope of the closing defense arguments, disclosure of the exact nature of the prosecution’s *891dealings with its key witness would certainly have affected the weight given his testimony by the jury.
Our alternative holding that Giglio requires new trials in some negligent nondisclosure situations where the showing of material harm falls short of the new trial standard in civil cases is also supported by a practical consideration. In cases such as this, a finding of intentional nondisclosure would carry with it a strong presumption that the prosecutor and members of his staff had perjured themselves. Faced with this prospect, trial judges might well be reluctant to find any nondisclosure intentional. Given the rigor of the legal standard for finding perjury, see Bronston v. United States, 409 U.S. 352, [93 S.Ct. 595, 34 L.Ed.2d 568] (1973), and the noncondu-civeness of a new trial motion procedural setting for getting all of the facts, findings of unintentional rather than intentional nondisclosure are likely to be more frequent.
There will likely be difficulty in distinguishing promises of relief and tacit understandings from assurances of lesser degrees of cooperation. Similarly, it may be difficult to .distinguish whether a “promise” of reduced sentence can be made by those who deal most directly with witnesses, since they will often lack formal authority to make such assurances.
Consequently, our reading of Giglio, requiring that negligent nondisclosure situations be assessed as to whether the jury’s earlier credibility assessment would have been affected by the newly disclosed evidence, appears a reasonable one. Due process rights would be preserved, yet new trials would not be required in every nondisclosure situation. In those uncommon cases where prosecu-torial misconduct was admitted or virtually uncontroverted, we would preserve an almost per se standard, granting new trials if the suppressed evidence would have been “favorable and material” to the defense. In the more common case of unintentional nondisclosure, prosecutors would be encouraged to disclose their full course of dealings with witnesses under our rule requiring a new trial whenever nondisclosure might have affected the jury’s assessment of credibility.
Reversed and remanded for a new trial.
At the conclusion of the most recent evidentiary hearing, the district court found that no one from the United States Attorney’s office had made any direct promises to Durden. While this new factual finding is supported by the evidence and is not clearly erroneous, it does not affect the outcome of this case.
The district court found that Durden and the agents had failed to disclose to the jury in the Butler trial the assurances the agents had made to Durden in return for his agreement to cooperate with the prosecution. The prosecutor is responsible for the nondisclosure of assurances made to his principal witnesses even if such promises by other government agents were unknown to the prosecutor. Since the investigative officers are part of the prosecution, the taint on the trial is no less if they, rather than the prosecutor, were guilty of nondisclosure. Barbee v. Warden, 331 F.2d 842, 846 (4th Cir. 1964).
In denying the defendant’s motion for a new trial, the district court failed to apply the legal standard required by the above-quoted decision of December 18, 1974. The district court concluded that the nondisclosure by Durden and the agents would not have affected the verdict. That is not the correct standard.
When the government permits a witness to parade himself in false colors, and the truth is discovered, the motion for a new trial should be granted if the false testimony (or nondisclosure) could, in any reasonable likelihood, have affected the judgment of the jury. Napue v. Illinois, 360 U.S. 264, 271, 79 S.Ct. 1173, 3 L.Ed.2d 217 (1959). In this case, in 1974, we held that a new trial is required whenever nondisclosure might have affected the jury’s assessment of credibility. Here, the falsehood of Durden was egregious, and clearly could have affected the jury’s assessment of his testimony.
Reversed and remanded.
United States v. Ted Dean Butler, No. 74-2201, and United States v. Robert Butler, No. 74-2000; opinion filed December 18, 1974, and withdrawn April 24, 1975. The present appeal is prosecuted by Robert Butler alone.
The agents testified that they gave no promises of leniency to Durden, had never indicated to him the possibility of seeking dismissal or reduction of charges he was facing, and were aware of no plan to offer such assistance in return for Durden’s cooperation.
Appellants’ skepticism as to these disclaimers appears justified in light of the very lenient treatment accorded to Dur-den after his testimony. Undoubtedly as a result of his cooperation, the government dismissed his federal case, and interceded in his two state cases to obtain dismissal in one and probation in the other.
The court made several findings, stating at one point:
“I conclude that before trial there was a plan to recommend to higher authority substantial benefit to Durden in the form of at least a misdemeanor if he fully cooperated.”
“His mention of dismissal as to Mr. Bernard and Mr. Russell may not have been in the form of a positive declaration that the case would be dismissed But the discussions could be in such a manner as to give the defendant’s counsel the idea that dismissal was in the cards.
“You remember to Mr. Bain he said that the case would probably be dismissed. Now that’s not saying a case is going to be dismissed. He said it would probably be dismissed. It leaves the inference, of course, that *888defense counsel might say, ‘Well, we were promised that this was going to be dismissed.’ ”
In this regard, the court noted that:
“The Government, through its agents and counsel, we know from experience that the Government, the Government agents and counsel for the Government have ways of indicating to defendant’s counsel that benefits are likely to result from cooperation. That can be indicated without making a bald promise that the charge is going to be reduced or that the case is going to be dismissed.”
See also note 3, supra.