Today we interpret a standard form Bankers Blanket Bond, in deciding an appeal from summary judgment granted to First National Bank of Louisville. The district court held that a Bond issued by Aet-na Casualty and Surety Co. and Federal Insurance Co. covered the dishonest or fraudulent acts of FNBL employee Kevin DeWitt, without proof that he intended by his fraudulent acts to cause any loss to the bank. We conclude that the bond insures only those fraudulent acts intended to cause a loss and hold that there are genuine issues of material fact as to the Bond’s coverage. We hold that there is a genuine issue of fact whether the employee on these facts intended to cause the bank a loss within the meaning of the bond. Finally, we conclude that the district court erred in instructing the jury that the definition of fraud in the insuring provision controls the definition of fraud in the termination provision. We reverse summary judgment, and remand for trial.
I.
The Sureties issued FNBL a standard form 1980 Bankers Blanket Bond to cover the fraudulent and dishonest acts of its employees for losses discovered from July 1, 1985 to July 1, 1986. In 1986, FNBL claimed coverage under Insuring Agreement (A) of the Bond for over $20 million in losses from eight failed loans caused by the dishonest acts of a young FNBL loan officer, Kevin DeWitt. Insuring Agreement (A) provides coverage under the bond for losses resulting directly from dishonest or fraudulent acts of an Employee committed alone or in collusion with others. Dishonest and fraudulent acts are defined for purposes of the insuring agreement as limited to only those acts “committed by such Employee with the manifest intent (a) to cause the Insured to sustain such loss by such Employee and (b) to obtain financial benefit for the Employee or any other person or organization intended by the Employee to receive such benefit other than salaries, commissions, fees, bonuses, promotions, awards, profit sharing, pensions, or other employee benefits earned in the normal course of employment.”
DeWitt joined FNBL’s management training program on his graduation from college in 1983. After his training, he was assigned to the Construction Loan Division where he prepared credit approval applications and administered the closing and payout of large commercial real estate loans. The eight loans at issue were construction loans presented by DeWitt for approval between 1984 and 1986, ranging from $1.9 *1164million to $10.9 million. DeWitt misrepresented several facts about these loans, most often falsifying the credit records of borrowers and guarantors. There was also evidence that DeWitt misrepresented the identity of permanent lenders, the number of pre-sold project condominiums, exaggerated the size of borrower’s contributions, and falsified various loan-related documents.
The Sureties presented evidence that FNBL learned that DeWitt had made misrepresentations and acted beyond the scope of his authority well before he made some of the failed loans. While DeWitt was in the training program his supervisors discovered that he was changing numbers in his credit analyses to make them balance, a technique referred to as “plugging.” Whether plugging is considered a dishonest practice is disputed. Although DeWitt was mildly reprimanded, the head of the CLD was not informed of this incident. In June 1985, a loan reviewer discovered major deficiencies in one of DeWitt’s loans. After meeting with DeWitt, the reviewer informed FNBL officials that DeWitt had distorted the credit history of a guarantor and did not make the loan to a borrower in the original write-up.
FNBL officials were also told that two of the failed loans had been closed “outside of approval,” meaning that the established .procedures for approval were not followed. In November 1985, DeWitt allegedly lied about whether the “Odyssey Two” loans had been approved by the Executive Loan Committee and was caught by an FNBL official, Despite these incidents, DeWitt was recommended for a bonus, promoted, and praised for his “extraordinary” results. The. Sureties assert that FNBL’s knowledge of these incidents triggered the Bond’s termination clause, ending coverage for DeWitt’s dishonesty.
In March 1986, DeWitt resigned his position at FNBL to accept an offer at a Baltimore bank. Nine days later he received a $40,000 “loan” from Loretta Lustig who was either individually or through a corporate entity the borrower or guarantor on a number of the failed loans. DeWitt claimed that the loan was intended to evidence Lustig’s good faith in her promise to later employ DeWitt.
Soon after DeWitt’s departure, serious problems with his loans began to emerge. On March 28, 1986, DeWitt returned to Louisville to meet with FNBL officials about their investigation of the credit information he had supplied for loans. DeWitt confessed to fabricating credit references. In a handwritten statement after the meeting DeWitt' stated that he falsified the information to make good loans and get recognition at FNBL, not to get personal gain from customers.
On April 2, 1986, FNBL gave the Sureties notice of a possible claim under the bond. The real estate markets in the states where most of the properties were located went into severe recession and none of the projects earned enough to pay the loans from FNBL. The losses on the loans totaled more than $20 million.
In July 1988, FNBL added the Sureties to its complaint previously filed against DeWitt and other parties. The Sureties denied liability under the Bond. In January 1989, DeWitt pleaded guilty to violating 18 U.S.C. § 1005. The guiíty plea was conditioned upon the district court’s acceptance of a sentencing agreement. The district court rejected that agreement as being too lenient and DeWitt withdrew his guilty plea. The district court granted partial summary judgment for FNBL, finding that DeWitt’s withdrawn guilty plea to a violation of 18 U.S.C. § 1005 in January 1989 satisfied the requirements for coverage under the fidelity bond. The district court also denied summary judgment to the Sureties and struck the Sureties’ defense that the losses were caused by a declining “oil patch” economy rather than directly by De-Witt’s conduct. The remaining issues were tried to a jury. The jury rejected the Sureties’ defense of termination and lack of timely notice. After the verdict for FNBL, the district court entered judgment in the amount of $20 million, the policy limit.
While this case was pending on appeal, DeWitt pled guilty to an 18 U.S.C. § 1005 offense under a plea agreement which was accepted by the district court. FNBL has *1165requested that this court take judicial notice of his second guilty plea -and statements made by DeWitt during his plea hearing.
II.
To recover under Insuring Agreement (A) FNBL must prove a loss resulting directly from dishonest or fraudulent acts by a bank employee committed with the manifest intent to cause a loss to the bank and the manifest intent to benefit the employee or someone else. The district court granted summary judgment on all the coverage issues for FNBL, holding that “As a matter of law, DeWitt’s withdrawn guilty plea in the Western District of Kentucky plainly satisfies the requirements for coverage under the fidelity bond.” In response to the Sureties’ motion for clarification, the district court again stated that all the coverage elements were found implicitly in the guilty plea adding simply that the elements are “further substantiated” by FNBL’s documentary evidence, including DeWitt’s confession, his default judgment in FNBL’s civil action against him, and other documents.
We conclude that the district court’s reliance upon DeWitt’s withdrawn guilty plea as conclusive proof of coverage under Insuring Agreement (A) of the policy was error. The statute under which DeWitt was attempting to plead guilty, 18 U.S.C. § 1005, criminalizes a number of deceptive or fraudulent practices. Although a guilty plea under 18 U.S.C. § 1005 might prove some of the elements required for policy coverage, it does not necessarily do so. Section 1005 does criminalize false entries in bank records which are made “with intent to injure or defraud such bank.” However, the same section also criminalizes false entries made “with intent ... to deceive any officer of such bank.”
FNBL would have us rely upon the statements made by DeWitt at his second guilty plea hearing which occurred after the summary judgment was granted by the district court. We restrict our review on summary judgment to materials which were before the district court. Nissho-Iwai American Corp. v. Kline, 845 F.2d 1300, 1307 (5th Cir.1988). However, even if we were to consider DeWitt’s second guilty plea, the fact of that plea alone would still not entitle FNBL to summary judgment in the face of countervailing evidence on the elements of policy coverage. A guilty plea to a criminal charge is evidence, perhaps powerful, but it is far from the only evidence of DeWitt’s intent presented in this case. In determining whether there is a genuine issue of material fact on each element of coverage, the guilty pleas should receive the same treatment as any other piece of evidence. We turn now to the question whether, on the basis of the entire record, summary judgment is appropriate on any of the elements of coverage.
A.
The first element of policy coverage is that the employee have a manifest intent to cause the bank a loss. A bank employee may commit dishonest or fraudulent acts with an intent to harm the bank or with an intent to benefit it. In some cases the nature of the dishonest act itself demonstrates the employee’s intent. At one end of the continuum are acts of embezzlement. FDIC v. St. Paul Fire and Marine Ins. Co., 942 F.2d 1032 (6th Cir.1991); Glusband v. Fittin Cunningham & Lauzon, 892 F.2d 208, 212 (2d Cir.1989) (limiting bond coverage to embezzlement or “embezzlement-like” acts). An embezzler always has a manifest intent to cause the bank a loss because his gains come only at the bank’s expense.
At the other end of the continuum are dishonest acts by employees which can only benefit the employee when the bank also benefits. In Municipal Securities v. Insurance Co. of N.A., 829 F.2d 7 (6th Cir.1987), for example, a trader intentionally violated her employer’s limit on the amount of government bonds which could be held in the company’s account at. any given time. To cover her losses she lied to her employer and falsely documented sales that had never taken place. The Sixth Circuit found that the employee’s motive was not to cause her employer a loss relying to a large extent on the fact that the employ*1166ee would benefit only if her employer also benefitted.
The fraudulent activities of DeWitt do not lie at either of these two poles, but lie somewhere in the middle of the continuum. In these situations intent becomes a question of fact which will generally not be subject to summary judgment. St. Paul Fire and Marine, 942 F.2d at 1036. De-Witt lied to obtain the bank’s approval on loans of higher risk than it would have approved without his fraudulent statements. Our question here is under what circumstances an employee who acts dishonestly to obtain the approval of loans has the manifest intent to cause the bank a loss.
This circuit has never addressed directly the question of when an employee who fraudulently obtains loan approval demonstrates manifest intent as defined in the 1980 bond. In its order clarifying its summary judgment, the district court relied upon Merchants and Farmers State Bank v. Fidelity and Casualty Corp., 791 F.2d 1141 (5th Cir.1981) to conclude that a dishonest act resulting in a risky loan being made is inevitably with manifest intent under Insuring Agreement (A) of the Bond. The bond in Merchants was issued in 19Í5, well before the adoption of the standard language being interpreted in this case. For that reason, the opinion refers to a quite different manifest intent than that required by the 1980 Bond. Merchants holds that the jury could have found “manifest intent to deceive the bank causing loss” from an employee’s dishonesty in falsely asserting that a guaranty from a creditworthy party had been obtained to support a risky loan. Id. at 1149. The context of this language makes clear that it was enough that the employee had a manifest intent to deceive, even if there was no intent to cause a loss. Thus, Merchants does not answer our question.
An employee may use fraudulent documents for loans, believing that they would be successfully paid, without manifest intent to cause the bank a loss. In First Federal Savings & Loan v. Transamerica, 935 F.2d 1164 (10th Cir.1991), for example, the court affirmed summary judgment for the insurer because the bank could not meet the manifest intent requirements of the Bond. An employee of the Bank fraudulently arranged loans after an initial application had been turned down. The court found that the employee’s, poor judgment in making these loans was not evidence of his manifest intent to cause the bank a loss, however likely such a loss may be where the undisputed evidence demonstrated that the employee hoped that the loans would benefit both the bank and the borrower.
To determine intent to cause a loss we do not inquire solely into the subjective motive or purpose of the employee. See St. Paul Fire and Marine, 942 F.2d at 1035 (courts must rely upon inferences from tangible manifestations of behavior, not merely upon actor’s subjective mental state to determine intent). Thus, the claim by an errant employee that no loss to the bank was intended will seldom be conclusive. When an employee obtains fraudulent loans with reckless disregard for a substantial risk of loss to the bank, a jury may infer from his reckless conduct and surrounding circumstances that he intended to cause that loss. The bank need not produce any direct evidence of the employee’s intent, but may rebut an insurer's motion for summary judgment with circumstantial evidence from which a reasonable jury could infer the intent to cause a loss. United States v. Adamson, 700 F.2d 953, 965 (5th Cir.1983) (en banc). (In criminal bank fraud prosecution, trier of fact may infer intent from reckless disregard of bank’s interest). The jury should be instructed that in answering the question of intended loss it should consider the range of evidentiary circumstances, including the relationship between the borrowers and the employee, the employee’s knowledge of the likelihood that the loans would not be repaid, and all the other surrounding circumstances bearing on the employee’s purpose.
We are persuaded that FNBL has not demonstrated that it is entitled to summary judgment. DeWitt’s withdrawn guilty plea to an 18 U.S.C. § 1005 offense is not conclusive, as we explained. In sum, the other evidence of DeWitt’s intent is mixed. In his original statement to bank officials he asserted that fraudulent acts were for the *1167purpose of getting ahead at the bank. His motives in making statements in connection with his guilty pleas which profess an intent to injure the bank are open to serious question. However, he left the bank before his fraud was discovered, and his relationship with Loretta Lustig outside the normal channels of his bank duties raises questions about his motives in obtaining these loans.
That is not to say that the Sureties are entitled to summary judgment on the issue of manifest intent. In this case, unlike Transamerica and Municipal Securities, there is significant evidence from which a jury could infer that DeWitt did have a manifest intent to cause FNBL a loss. This element of coverage is a disputed issue of material fact which should have been left to the jury.
B.
The district court also erred in granting summary judgment on the issue of DeWitt’s manifest intent to benefit himself (outside the normal salaries, bonuses, promotions, etc.) or others by his conduct. There is evidence from which a reasonable jury could infer that DeWitt intended to benefit himself by benefitting Loretta Lus-tig, who was involved in the projects receiving between 2 and 6 of the disputed loans. Among other things, there was evidence that when DeWitt left FNBL Lustig gave him a $40,000 “loan,” ostensibly to show her good faith in an offer of future employment to him.
This evidence is not uncontradicted, however. In his original statements to bank officials and other evidence, DeWitt indicated that he falsified the loan documents to get ahead at the bank. The record indicates that he was successful at the bank, being recommended for raises and promotions based upon his loan production. A jury could infer that DeWitt acted with the intent to benefit himself at the bank through career advancement and not for direct personal financial gain. Thus, the question of DeWitt’s manifest intent to benefit himself or another was for the jury.
C.
The Sureties argue that the district court erred in granting summary judgment on the issue of whether FNBL’s losses resulted directly from DeWitt’s fraudulent conduct. The district court stated only that the causation requirement is “satisfied implicitly when DeWitt’s withdrawn guilty plea is read in conjunction with the Bond.” We find that whether the loan committee relied upon DeWitt’s misrepresentations in making at least some of the disputed loans is a disputed issue of material fact. An expert witness for the Sureties testified that the mistakes and inconsistencies in some of the reports were so obvious that any experienced banker would have recognized them if he had read the report. They also argue that in many cases the borrower’s actual credit report was at least as good as the one DeWitt fabricated. Also, on at least one loan the report was dated after the Loan Committee had approved the loan. This evidence raises an issue of fact as to whether FNBL would have approved each of these loans without the dishonest acts of DeWitt.
The Sureties also argue that the district court erred in striking their “oil patch” defense. We conclude that the district court was'correct in holding that the decline in the general real estate market in the area does not preclude a finding of causation under the Bond. The Sureties would have us read the requirement that the loss be directly caused by the dishonest or fraudulent act narrowly. Such a reading would, however, all but eliminate coverage for loans made because of dishonest or fraudulent acts. There will always be some intervening cause for the failure of these loans to be repaid; otherwise the bank would suffer no loss. See Mitsui Manufacturers Bank v. Federal Ins. Co., 795 F.2d 827, 831 (9th Cir.1986) (rejecting a policy interpretation which would allow the cause to always be attributed to whatever circumstances made a deposit uncollectible). A loss is directly caused by the dishonest or fraudulent act within the meaning of the Bond where the bank can demonstrate that it would not have made the loan in the absence of the fraud. The Bond *1168does not require the bank to rule out all reasons the loan was not repaid before it can obtain coverage.
III.
The Sureties argue that the district court erred in instructing the jury whether FNBL’s knowledge of dishonest behavior by DeWitt terminated coverage for losses he thereafter caused. The district court instructed the jury that, for purposes of the termination clause, dishonest or fraudulent acts are “any deceitful act committed by Kevin DeWitt with the manifest intent to cause the bank to sustain a loss and to obtain financial benefit for himself or for any other person or organization.” This instruction incorporates the restrictive definition of dishonest or fraudulent acts from Insuring Agreement (A) into the termination clause. We agree with the Sureties that the jury charge misinterpreted the operation of the termination clause.
Two aspects of the Bond language convince us. First, the restrictive definition of dishonest and fraudulent acts in Insuring Agreement A contains the proviso “as used in this Insuring Agreement.” The insuring agreements are specifically identified portions of the overall Bond. This definition is in contrast to the general definitions in the Bond which are preceded by the language “as used in this bond.” There is no general definition of dishonest or fraudulent acts.
Second, the termination clause language itself is inconsistent with the restrictive definition of dishonest and fraudulent acts used by the district court. The termination clause terminates coverage when the insured learns that the employee has committed “any dishonest or fraudulent act ... against the Insured or any other person or entity.” Requiring that the dishonest act be done with manifest intent to cause the bank a loss effectively eliminates the terminating effect of dishonest acts against other persons or entities. Thus, the reading of the termination clause asserted by FNBL and adopted by the district court makes an entire phrase of the clause superfluous.
FNBL argues that because the only losses caused by employees which are referred to in the termination clause and covered by the Bond are those covered by Insuring Agreement A, the definition of dishonest and fraudulent acts in both provisions must be the same. They also argue that the definition of dishonest and fraudulent acts in Insuring Agreement (A) does not confine itself to that Insuring Agreement because it does not say “as used in this Insuring Agreement only.” FNBL responds to the argument that its interpretation of the termination clause fails to accommodate the language pertaining to acts against another person or entity by positing an act against another individual for which the bank would be vicariously liable.
We find FNBL’s arguments unconvincing. The only reasonable interpretation of the termination clause is that dishonest acts short of those triggering coverage under Insuring Agreement A may terminate coverage under the Bond. We do not believe that the termination clause was intended to restrict the bank’s duty of care so severely. Moreover, the language of the termination provision envisions termination for dishonest acts committed against other entities. The district court erred in its jury instruction on this issue.
On remand, the termination issue should be submitted to the jury without the restrictive definition of dishonest or fraudulent conduct contained in Insuring Agreement (A).
IV.'
The Sureties also assert that the district court erred in instructing the jury that the Sureties bore the burden of proof on the issue of timely notice, relying upon Reserve Insurance Co. v. Richards, 577 S.W.2d 417, 419 (Ky.1979). Since the trial of this case, the Kentucky Supreme Court has revisited this issue and overruled Reserve Insurance in Jones v. Bituminous Casualty Corp., 821 S.W.2d 798 (Ky.1991). Under Jones, the insurer has the burden of proving probable prejudice as a result of a delay in notice by the insured. Id. On remand, the district court must determine the proper application of Jones to the trial of this case.
*1169V.
In motions filed with this court after oral argument, the Sureties have alleged impropriety on the part of FNBL and the district court in the Western District of Kentucky which accepted DeWitt’s guilty plea. They asked us to stay proceedings and to prohibit the use of DeWitt’s guilty pleas. As we explained, the guilty plea- is not controlling but is, rather, an evidentiary circumstance to be considered with all the other facts of the case. We do not decide the admissibility of the plea and the events surrounding it. We decide only that, if admissible, it will not be controlling. The motions filed may affect the admissibility of the plea and the statements made in that proceeding and other facts relevant to the case may be developed. These questions now swirling around the plea are best handled by the trial court by an in limine hearing or otherwise, as the court in its discretion may choose.
VI.
On cross appeal, the Bank argues that the District Court erred in failing to award pre-judgment interest, ruling that the policy limit is $20 million, and applying post-default recoveries to the principal of the loan. We decline to address these issues in this appeal because the necessity of resolving them depends upon the result on remand. For the foregoing reasons, we REVERSE the judgment of the district court and remand for proceedings consistent with this opinion.