Appellant Kevin T. Dowling sued Appel-lee American Hawaii Cruises, Inc. (“AHC”), under the Jones Act, 46 U.S.C. § 688, and general maritime law, for injuries he suffered while working aboard AHC’s ship the S.S. Independence. The case went to a jury, which found against Dowling on the grounds that AHC was not negligent and that the S.S. Independence was seaworthy. Dowling claims that the district court erred both by finding that the minutes of meetings of the ship’s safety committee were privileged and therefore immune from discovery, and by applying the wrong standard for seaworthiness in deciding Dowling’s motion for a new trial. The district court had jurisdiction over Dowling’s claims under 28 U.S.C. §§ 1331 and 1333. We have jurisdiction pursuant to 28 U.S.C. § 1291. Because we hold that the district court erred in concluding that the safety committee minutes were immune from discovery, and therefore grant Dowling a new trial, we need not consider whether the district court properly defined the law of seaworthiness in denying Dowl-ing’s motion for a new trial.

I

FACTS

Dowling was First Mate aboard the S.S. Independence, a cruise ship that makes voyages around the Hawaiian Islands. On December 1, 1988, while supervising mooring operations, Dowling slipped and fell onto the deck of the S.S. Independence, seriously injuring his back. He alleges that he slipped on oil leaking from a defective “roller chock,” which is a device that prevents mooring lines from chafing against the bulkheads and the hull of the ship. On August 22, 1989, Dowling filed suit against AHC alleging that AHC was negligent in not repairing the roller chock, and that the S.S. Independence was unsea-worthy.

On January 10, 1990, Dowling served AHC with his first set of interrogatories. Interrogatory 29 requested: “Describe your procedures regarding the safety of employees, i.e., officers and crew on board the S.S. Independence.” Other interrogatories specifically requested information regarding the roller chock. AHC refused to answer Interrogatory 29, and also refused to turn over the minutes of the meetings of the ship’s “safety committee” held during the period from November 12, 1987 to December 1, 1988. AHC asserted a privilege of “self-critical” analysis, which it alleged protected this information from discovery. Nonetheless, AHC ultimately produced heavily redacted excerpts of the minutes from four safety committee meetings in 1987. The excerpts contained references to the leaking roller chock, and AHC claimed they were the only minutes relevant to Dowling’s claim.

On February 28, 1990, Dowling filed a motion to compel AHC to answer his interrogatories. The magistrate denied the motion, treating it as one to compel production of the safety committee minutes and finding that the asserted privilege of self-critical analysis shielded the minutes from discovery. AHC cited the order in refusing to[*425] answer interrogatories regarding who investigated or repaired the roller chock leaks, how the bow area was kept free of grease, and who set the agenda for the safety committee meetings. Captain Lawrence Kelly and Chief Mate David Sloane relied on the order in refusing to answer deposition questions regarding who knew what about the leaking roller chock.

Dowling appealed the magistrate’s order to the district court. The court held that the minutes, with the exception of the excerpts already produced, were covered by a privilege of self-critical analysis. See Dowling v. American Hawaii Cruises, Inc., 133 F.R.D. 150, 153-154 (D.Haw.1990). In its order denying Dowl-ing’s motion for clarification, the court confirmed that the privilege covered neither the voluntarily disclosed excerpts, nor questions to witnesses regarding any objective evidence relating to the roller chock. Dowling was not entitled, however, “to delve into the minds of the safety committee members or the underlying operations of the safety committee.” The order denied Dowling’s request to certify the issue to the Ninth Circuit.

On October 3, 1990, the case went to trial, where AHC relied on the district court’s ruling to object to questions Dowl-ing put to witnesses regarding the procedures of the safety committee. Dowling did not pursue those questions. On October 19, 1990, the jury returned a verdict that AHC was not negligent and that the S.S. Independence was seaworthy.

II

THE PRIVILEGE OF SELF-CRITICAL ANALYSIS

We review a district court’s rulings concerning discovery for abuse of discretion. Ah Moo v. A.G. Becker Paribas, Inc., 857 F.2d 615, 619 (9th Cir.1988).

Rule 26(b)(1) of the Federal Rules of Civil Procedure provides that “[pjarties may obtain discovery regarding any matter, not privileged, which is relevant to the subject matter involved in the pending action.” Rule 501 of the Federal Rules of Evidence states that in cases in which federal law governs, privileges “shall be governed by the principles of the common law as they may be interpreted by the courts of the United States in the light of reason and experience.” Rule 501 reflects Congress’s intent to allow courts “flexibility to develop rules of privilege on a case by case basis.” University of Pa. v. EEOC, 493 U.S. 182, 189, 110 S.Ct. 577, 582, 107 L.Ed.2d 571 (1990). The Supreme Court has recently stated that it is “disinclined to exercise this authority expansively.” Id. The Court has reiterated its long standing view that the policy favoring open discovery requires that privileges must be “strictly construed.” Id.

This circuit has not yet considered whether there exists a so-called privilege of self-critical analysis, ^[1] but other courts have generally required that the party asserting the privilege demonstrate that the material to be protected satisfies at least three crite[*426] ria: “first, the information must result from a critical self-analysis undertaken by the party seeking protection; second, the public must have a strong interest in preserving the free flow of the type of information sought; finally, the information must be of the type whose flow would be curtailed if discovery were allowed.” Note, The Privilege of Self-Critical Analysis, 96 Harv.L.Rev. 1083,1086 (1983). To these requirements should be added the general proviso that no document will be accorded a privilege unless it was prepared with the expectation that it would be kept confidential, and has in fact been kept confidential. See James F. Flanagan, Rejecting a General Privilege for Self-Critical Analyses, 51 Geo.Wash.L.Rev. 551, 574-76 (1983) (citing 8 J. Wigmore, Wigmore on Evidence § 2285, at 527 (1961)); see also Peterson v. Chesapeake & Ohio Ry., 112 F.R.D. 360, 363 (W.D.Mich.1986) (refusing to apply the privilege to investigative report because report was not “performed with the expectation that the analysis [would] remain confidential” and in fact had not been kept confidential); Westmoreland v. CBS, Inc., 97 F.R.D. 703, 706 (S.D.N.Y.1983) (same).

In this case, the district court did not attempt to determine whether the safety committee minutes satisfied any set of criteria, but rather concluded that the “essential test” of whether the privilege applies “involves balancing the public interest protected by the privilege against the plaintiff’s need for the material to make his case.” Dowling, 133 F.R.D. at 153. It concluded that the alleged “chilling effect” disclosure would have on the candid assessment of safety issues outweighed Dowl-ing’s need for the minutes because Dowl-ing had obtained the information he needed relating to the roller chock. We believe the district court struck the wrong balance, and we hold that no privilege of “self-critical analysis” protects routine internal corporate reviews of matters related to safety concerns.

Even if such a privilege exists, the justifications for it do not support its application to voluntary routine safety reviews.

First, such reviews will rarely, if ever, be curtailed simply because they may be subject to discovery. Organizations have many incentives to conduct such reviews that outweigh the harm that might result from disclosure. The most prominent of these is surely the desire to avoid law suits arising from unsafe conditions. But organizations also have a strong incentive to avoid developing a reputation for having an unsafe premises. Such a reputation would make it more difficult for a corporation to attract desirable employees, and in the case of a cruise ship, a hotel, or a restaurant, to attract customers. One need only view a few automobile advertisements to recognize that manufacturers perform safety tests not required by law not only because of the threat of products liability suits for design defects, but because a reputation for safety renders a product more marketable. It is noteworthy that automobile and other manufacturers continue to conduct such reviews despite the infamous cost-feasibility memorandum that subjected Ford Motor Company to enormous punitive damages in Grimshaw v. Ford Motor Co., 119 Cal.App.3d 757, 174 Cal.Rptr. 348 (1981). See David G. Owen, Problems in Assessing Punitive Damages Against Manufacturers of Defective Products, 49 U.Chi.L.Rev. 1, 16-19 (1982).

Second, we do not expect that such reviews are always performed with the expectation that they will be kept confidential. Despite AHC’s assertions — unsupported by evidence — that its reviews were prepared with such an expectation, we suspect that reviews of the safety conditions of areas in which employees work are typically not confidential. We find it difficult to imagine what the point would be of gathering information regarding safety hazards on ship and then keeping that information secret.

Finally, the fairness rationale offered to justify application of the privilege to documents that a party has been legally required to prepare is inapplicable to voluntarily conducted safety reviews. It may be unfair for a court to require a party to turn over to an opposing litigant self-damning[*427] assessments that the government has required it to prepare. See O’Connor v. Chrysler Corp., 86 F.R.D. 211, 218 (D.Mass.1980). But this concern obviously does not exist when the party has engaged in the self-evaluation voluntarily. See Resnick v. American Dental Ass’n, 95 F.R.D. 372, 375 (N.D.Ill.1982).

On the other hand, information regarding a company’s response to the particular hazard, as well as its general safety procedures, will be invaluable to a plaintiff attempting to prove that his injury was caused by the company’s negligent failure to make safe a hazardous condition. That information may reveal not only what the company knew of the condition, but whether it regarded it as dangerous, whether it assigned someone to repair the condition, what that person did, whether the company typically responded swiftly and effectively to hazardous conditions or allowed them to persist, and so on. Furthermore, the implications of such a rule would have major ramifications outside the area of negligence. A rule that manufacturers’ reviews of the safety of their products are privileged from discovery would place a nearly insurmountable barrier in front of plaintiffs who must prove malice in order to obtain punitive damages in product liability actions. See Owen, supra, at 16-19. It would also create an enormous obstacle to claims that a manufacturer fraudulently concealed its knowledge of the health hazards of its products. See, e.g., Cipollone v. Liggett Group, Inc., — U.S. -, 112 S.Ct. 2608, 120 L.Ed.2d 407 (1992) (alleging cigarette manufacturer’s fraudulent misrepresentation of health hazards created by cigarettes).

We do not include within the category of routine safety reviews investigations designed to root out the causes of an accident that has already occurred. The protection of a report generated by such an investigation was the issue in Granger v. National R.R. Passenger Corp., 116 F.R.D. 507 (E.D.Pa.1987). There, the court applied the privilege to portions of the report that “clearly encompass[ed] opinions and recommendations.” Id. at 510. It reasoned that “[tjhere is no question that the public has an interest in the institution of practices assuring safer operations of railroads. The production of these portions of the report would tend to hamper honest, candid self-evaluation geared toward the prevention of future accidents.” Id.

We do not disagree with this analysis, but we believe the difference between pre-accident safety reviews and post-accident investigations is an important one. The candid analysis of the causes of accidents is more likely to be stifled by a disclosure requirement than would the routine review of safety concerns. This is because pre-accident safety reviews are designed to preempt litigation; it is perverse to assume that the candid assessments necessary to prevent accidents will be inhibited by the fear that they could later be used as a weapon in hypothetical litigation they are supposed to prevent. The Federal Rules of Evidence recognize this, and therefore bar only evidence of subsequent remedial measures to prove negligence. See Fed. R.Evid. 407 and advisory committee’s note (rationale for exclusion “rests on a social policy of encouraging people to take, or at least not discouraging them from taking, steps in furtherance of added safety”).

We hold that voluntary routine preaccident safety reviews are not protected by a privilege of self-critical analysis, and that the district court therefore abused its discretion by applying a privilege of self-critical analysis to the safety-committee minutes that Dowling sought to discover. Because the minutes contained material potentially relevant to Dowling’s claim, the verdict for AHC is REVERSED and the case is REMANDED for a new trial. Prior to trial, the district court shall direct AHC to produce the safety committee minutes sought by Dowling, insofar as such production is consistent with Fed.R.Civ.P. 26(b)(1) and this opinion.