Jerome Andrulonis contracted the disease of rabies while conducting a laboratory experiment with a rabies viral strain. He became tragically ill and eventually suffered serious and permanent brain damage, leaving him with the mental capacity of a four-year-old. He also suffered marked personality and behavioral changes as a result of dementia secondary to rabies encephalitis. Andrulonis's mental incapacity and emotional instability are permanent. The experiment, which was conducted under the supervision of Dr. John G. Debbie, a scientist employed by the New York State Department of Health ("NYSDOH"), and observed by Dr. George M. Baer, a federal government scientist, took place in the state-operated Griffin Laboratories just outside Albany, New York.

Joanna Andrulonis, individually, and as conservator of the property of Jerome Andrulonis, her husband, brought this action against the United States of America and a number of nongovernmental defendants to recover damages for the severe injuries they have suffered because of the rabies. The claim against the United States was, of course, brought under the Federal Tort Claims Act, 28 U.S.C. Secs. 1346, 2671-2680 ("FTCA" or "the act"). The United States cross-claimed against the nongovernmental defendants and filed a third-party complaint for contribution against the State of New York. See N.Y. Civ.Prac.L. & R. Secs. 1401-1404. Before trial, the district court dismissed Joanna Andrulonis's individual claims against the United States for failure to timely file the administrative claim required by the FTCA. Andrulonis v. United States, No. 79-CV-847, slip op. at 7 (N.D.N.Y. March 8, 1984); see 28 U.S.C. Sec. 2675(a) and Sec. 2401(b).

Before the end of trial, plaintiffs, through a series of agreements, had settled with all of the nongovernmental parties for a total of $1,300,000. The settling parties agreed that $225,000 would be paid in satisfaction of the primary claims of Jerome Andrulonis and that $1,075,000 would be paid in satisfaction of the derivative claims of Joanna Andrulonis.

After a nonjury trial of the remaining unresolved FTCA claims of plaintiff Jerome Andrulonis against the United States and the contribution claims of the United States against NYSDOH, Judge Munson found the government liable in negligence for Jerome Andrulonis's injuries and also found NYSDOH liable to the government on its contribution claim. In addition, considering the relative culpabilities of the settling private defendants, he found Eli Lilly and Company and John L. Thompson & Sons and Company responsible to the extent of 5 percent and exonerated the other settling defendants. The court assessed damages in the amount of $6,424,641, including $2,417,238 for the future custodial care of Jerome Andrulonis, and apportioned them 30 percent to the United States, 65 percent to NYSDOH, and 5 percent to Lilly/Thompson. Andrulonis v. United States, 724 F. Supp. 1421 (N.D.N.Y.1989).

Under section 15-108 of New York's General Obligations Law, the United States was entitled, with respect to each settlement, to a setoff for the greater of "(a) the amount stipulated in the settlement agreement, (b) the consideration paid for the release, or (c) the amount of the settling defendant's equitable share of the damages." N.Y.Gen.Oblig.Law Sec. 15-108. As noted earlier, Jerome Andrulonis had been allocated $225,000 of the proceeds of a settlement agreement with the private defendants. Of this amount, Lilly/Thompson paid $100,000 and the other settling private defendants paid $125,000. Since defendants Lilly/Thompson had settled for an amount that was less than their equitable share, the district court reduced the award by 5 percent or $321,232, representing the proportionate culpability found against Lilly/Thompson. Since the district court found no culpable conduct attributable to the other settling defendants, it reduced Jerome Andrulonis's award by the actual dollar amount paid to him by those defendants--$125,000. This left the United States primarily responsible to Jerome Andrulonis for $5,978,409, subject to recovering from the state 68.42 percent ( 65/95ths) of that amount, or $4,090,427, on its third-party contribution claim. N.Y.Civ.Prac.L. & R. Sec. 1402.

Given the complexity of this case, we must delve with considerable detail into the circumstances surrounding Andrulonis's illness and the experiment that caused it. It is unfortunate that such a long and technical narrative must follow, but a full understanding of the district court's decision cannot be reached without careful consideration of the virus used, the disease it causes, and the methods of researching rabies used at the time the tragedy occurred. The following facts include those that the district court found rabies experts reasonably should have known about the disease. Since the district court's findings are not clearly erroneous, we must accept them as true. Fed.R.Civ.P. 52(a).

Jerome Andrulonis was a 34-year-old senior bacteriologist employed to conduct rabies research in NYSDOH's Griffin Laboratory under the supervision of Dr. John G. Debbie, a research scientist employed by NYSDOH as head of the rabies labs at Griffin. Dr. Debbie was persuaded to conduct rabies research by Dr. George M. Baer, Chief of the Viral Zoonosis Branch and Rabies Laboratory at the Center for Disease Control ("CDC"), a division of the United States Department of Health and Human Services. The research being done by Dr. Debbie and Andrulonis was part of a joint effort by the NYSDOH and the CDC to develop a technique for mass immunization of wildlife in the hope of eradicating the incidence of rabies transmitted to man.

To realize this goal, Dr. Debbie was working on an oral vaccine that could be distributed to wildlife. Because the initial virus strains tested by Dr. Debbie could not withstand the destructive stomach acids of some species, he decided, upon the advice of Dr. Baer, to try a method called "enteric coating." Under this method, a pill, or nonpareil, would be coated first with the rabies virus, then a protective coating, and finally, a shell that could withstand stomach acids but would disintegrate in the animal's intestines, allowing the virus to be absorbed there. The virus would not cause development of the disease, but would cause the body to produce antibodies that would protect the animal against subsequent exposures to the rabies virus.

Dr. Debbie contacted the coating department at defendant WARF Institute, Inc. ("WARF"), which on March 30, 1976, sent a representative to demonstrate a coating technique using a Uni-Glatt machine, manufactured by defendant Glatt GmbH and distributed by defendant Glatt Air Techniques, Inc. ("the Glatt defendants"). The Uni-Glatt machine operated by suspending nonpareils in a column of upflowing air confined in a tube. The suspended nonpareils would then be coated with aerosolized virus.

When the machine operated, the aerosol itself could be seen inside the tube, although any leakage into the atmosphere was not visible to the unaided eye and left no trace outside the machine. However, nonpareils would occasionally leak out of the machine into the lab environment, and the final coating process created a visible reddish dust that escaped from the machine leaving the entire laboratory coated with the reddish dust after each run. During the initial demonstration run, it became apparent to everyone in the laboratory that the machine was not airtight. Despite this, the only precautions required of the lab workers were the wearing of a mask and gloves, and the maintenance of a high level of rabies antibodies in their blood systems through periodic inoculations with a rabies vaccine manufactured by defendant Eli Lilly and Company and distributed by defendant John L. Thompson & Sons and Company ("Lilly/Thompson").

Funding for the Uni-Glatt rabies experiments was provided by the federal CDC. In exchange, Dr. Debbie was to provide the CDC with some of the coated nonpareils prepared in the course of the Uni-Glatt experiments. Beyond providing the funding, CDC officials made no attempt to direct the day-to-day operations of the project, to devise any protocol, or otherwise to control the research efforts at Griffin.

Dr. Debbie used the Uni-Glatt in two more experiments in July and September of 1976, both times using a rabies viral strain prepared by Andrulonis with a titer of 6. "Titer" is the term used to indicate a standardized measure of a substance's ability to produce a defined result. With respect to the rabies virus, the titer measures the ability to kill laboratory mice after intracerebral injection of a solution containing that virus. The titer is expressed exponentially.

During the July and September runs, heat generated by the machine killed most of the viral particles and reduced the titer of the vaccine to 1, a level that produced irregular serum antibody levels. To overcome this problem, Dr. Debbie asked Dr. Baer to prepare a viral strain with a higher titer.

Dr. Baer, in keeping with CDC's general practice of preparing and supplying viral strains to many private and state-operated labs, prepared a relatively unknown viral strain, ERA-BHK/21, with a titer of 8.1. This was derived from a commercially available ERA vaccine, but it had a much higher concentration of viral particles and was less susceptible to heat. While the basic ERA vaccine had several known favorable characteristics, the particular characteristics of ERA-BHK/21 were relatively unknown to most scientists in March of 1977, since it was several times removed from the commercial ERA vaccine. ERA-BHK/21 was created through a process called "passaging" that had the potential for producing unpredictable changes in characteristics. The handful of studies on ERA-BHK/21 had reported mixed results about pathogenicity, and no one knew with certainty whether man was susceptible to developing the disease of rabies as a result of infection from this ERA-derived virus strain. The CDC generally recommended, as a precaution, that the virus be treated as pathogenic to man.

Several other factors, in addition to uncertainty about the characteristics of ERA-BHK/21, complicated the situation. People usually contract rabies through the saliva of an infected animal during a bite. Infection through aerosol exposure, the route of infection in Andrulonis's case, is much more rare and was not even recognized until the late 1950s or early 1960s. Not all exposures lead to infection; not all infections lead to the disease. Regardless of the route of infection, the chances of getting the disease after exposure are closely related to the quantity of viral particles, or the dose, to which a person is exposed. When the exposure is aerosol in nature, the dose is probably the single most important factor in determining whether the virus will cause the disease.

In addition, the pathogenesis of the disease is very different between the airborne and the bite routes. With a bite, the area is bathed with blood, so if the bitten person has been immunized, the antibodies in the blood have an opportunity to neutralize the virus before it can invade the central nervous system. In contrast, when airborne, the virus hits olfactory nerves that are exposed to the environment and that have as their only protection the nasal mucosa. Ordinarily, olfactory nerves do not have significant contact with antibodies in the blood stream. Thus, even if a person has been immunized, the antibodies do not have a chance to come into contact with the infected nerves, unless the vaccine is able to produce antibodies in the mucosa.

In 1976-1977, when Andrulonis was exposed to the virus, it was not known whether the Lilly vaccine he and the other lab technicians were using could protect against aerosol exposure. Lilly did not conduct studies even after becoming aware of the possibility of aerosol exposure, nor did it warn users of the potential that its vaccine might not protect against aerosol exposures. In light of the meager then-existing knowledge about aerosol exposures and vaccines, the district court found that safe lab practices mandated avoidance of all aerosol exposures and that virus-laden aerosols should not even be created unless absolutely necessary for the research goal, and in those cases, the aerosols should be physically contained using one of the readily-available containment systems.

On the evening of March 28, 1977, Dr. Baer personally delivered the specially prepared ERA-BHK/21 virus strain with a titer of 8.1 to Dr. Debbie at Griffin. He informed Dr. Debbie of the titer and Dr. Debbie relayed this information to the other lab workers the next morning. No disclosure was made about how the strain was prepared or how many times it was passaged. No safety inspection was made of the lab premises, but this was consistent with ordinary CDC practice in similar circumstances.

The tragic experiment took place the next day, March 29th. With Dr. Baer observing, Dr. Debbie started the machine and began circulating the nonpareils in the glass tube. The aerosolization process using the ERA-BHK/21 virus lasted over an hour, during which time both Dr. Debbie and Andrulonis worked in close proximity to the machine. The process required close attention, and usually Andrulonis or Dr. Debbie sat near the machine and tapped the glass tube with the handle of a screwdriver to prevent nonpareils from sticking together. Commonly, Andrulonis would place his face within six to twelve inches of the tube while tapping it, and would remain that close to the machine for minutes at a time throughout the run. Dr. Baer also observed the process closely and, at times, was within a foot of the machine.

Dr. Baer left on the evening of March 29th. Another coating run was done the next day, but this time using a viral strain prepared by Andrulonis that had a titer of 5.5.

The district court found on adequate, albeit disputed, evidence that Andrulonis had contracted rabies during the March 29, 1977, experiment that Dr. Baer observed. To begin with, Andrulonis's only contact with the virus was at work. On the job, it was unlikely that he got rabies from routine procedures; more likely, he contracted the disease from an "unusual" exposure. Of his four possible exposures between March of 1976 and March of 1977, compelling evidence points to the March 29, 1977, experiment as the cause of his illness, which first appeared in April of 1977. As to the other exposures, in March, July and September of 1976, a six-month incubation period is within the realm of possibility, but the average is 20-60 days. Only 14 percent of the incidents of rabies have an incubation period greater than 90 days regardless of the method of exposure. Since this was an aerosol exposure, the incubation period was more likely to be shorter than longer.

Another unusual aspect of the March 29th experiment was its use of the most highly concentrated virus with a titer of 8.1. This means there were 100 times more viral particles than a 6.0 titer and 400 times more viral particles than a 5.5 titer. Further, the viral strain used in the March 29th experiment was more resistant to the heat generated by the machine. Finally, the district court accepted the opinion testimony of various experts who stated with a reasonable degree of medical and scientific certainty that Andrulonis contracted the disease of rabies as a result of exposure in the March 29th experiment.

The district court found the government liable under New York law on three independent theories of negligence: supplier's duty to warn, non-negligent creation of a hazard, and negligent entrustment. The government challenges each of these holdings. The district court also held that the discretionary function exception of the Federal Tort Claims Act was inapplicable to this case, and both the government and NYSDOH appeal from this holding.

In addition, the government and NYSDOH appeal from the district court's approval of the allocation of settlement proceeds; NYSDOH appeals the district court's apportionment of damages; and finally, Jerome Andrulonis cross-appeals from the court's determination of the cost of future custodial care.

We first turn to the question of whether the discretionary function exception protects the government from suit in these circumstances. If the exception were to apply in this case, it would bar this action against the government, and we could dismiss the complaint on that ground alone. However, since we determine that the exception is inapplicable here, we must then discuss the question of whether New York law imposes liability on the government under the circumstances of this case. Of the three theories relied on by the district court, we determine that the failure to furnish an adequate warning is sufficient to hold the government liable, so we need not reach the merits of the other two theories.

We next will address the questions of whether the settlement proceeds were properly allocated between Joanna and Jerome Andrulonis, and whether the district court properly apportioned the percentages of responsibility between the federal government and the state. Finally, we will discuss the plaintiff's cross-appeal on whether the district court correctly calculated his damages for future custodial care.

Initially, we take up whether this action was barred by the discretionary function exception to the FTCA. The tort claims act generally authorizes suits against the United States for damages

for injury or loss of property, or personal injury or death caused by the negligent or wrongful act or omission of any employee of the Government while acting within the scope of his office or employment, under circumstances where the United States, if a private person, would be liable to the claimant in accordance with the law of the place where the act or omission occurred.

28 U.S.C. Sec. 1346(b). The act further provides that the United States shall be liable with respect to tort claims "in the same manner and to the same extent as a private individual under like circumstances." 28 U.S.C. Sec. 2674.

Although the act waives the sovereign immunity of the United States, it does not do so in all respects. Congress carefully excepted several important classes of tort claims from the act's broad waiver of immunity. Of particular relevance here, section 2680(a) provides that the act shall not apply to

[a]ny claim based upon * * * the exercise or performance or the failure to exercise or perform a discretionary function or duty on the part of a federal agency or an employee of the Government, whether or not the discretion involved be abused.

28 U.S.C. Sec. 2680(a). "The discretionary function exception * * * marks the boundary between Congress' willingness to impose tort liability upon the United States and its desire to protect certain governmental activities from exposure to suit by private individuals." United States v. S.A. Empresa De Viacao Aerea Rio Grandense (Varig Airlines), 467 U.S. 797, 808, 104 S. Ct. 2755, 2761-2762, 81 L. Ed. 2d 660 rehearing denied, 468 U.S. 1226, 105 S. Ct. 26, 82 L. Ed. 2d 919 (1984).

The Supreme Court in Dalehite v. United States, 346 U.S. 15, 27-28, 73 S. Ct. 956, 963-964, 97 L. Ed. 1427 (1953), analyzed the legislative history of the act and concluded that its waiver of sovereign immunity was inapplicable to "acts of a governmental nature or function." The basis for the exception was congress's wish "to prevent judicial 'second-guessing' of legislative and administrative decisions grounded in social, economic, and political policy through the medium of an action in tort." Varig Airlines, 467 U.S. at 814, 104 S.Ct. at 2764-2765. The foundation for this exception is protection of "the principles embodied in the separation of powers doctrine by keeping the judiciary from deciding questions consigned to the executive and legislative branches of the government." Caban v. United States, 671 F.2d 1230, 1233 (2d Cir.1982).

By barring tort liability for activities that require the alleged tortfeasor to consider and weigh competing policies in arriving at his decision, this test protects courts from "involve[ment] in making ... decision[s] entrusted to other branches of the government," and especially from questions involving "not negligence but social wisdom, not due care but political practicability, not reasonableness but economic expediency".

Id. (citations omitted). Congress thus sought to assure that tort liability would not "seriously handicap efficient government operations." United States v. Muniz, 374 U.S. 150, 163, 83 S. Ct. 1850, 1858, 10 L. Ed. 2d 805 (1963).

Determining whether specific conduct is protected by the discretionary function exception to the FTCA is sometimes difficult. The act itself contains no definition of the types of discretionary functions within the exception. "The principal difficulty is simply that all federal employees exercise a certain amount of discretion in the discharge of their responsibilities." Caban, 671 F.2d at 1232.

To determine whether conduct falls within the exception, we must focus on "the nature of the conduct, rather than the status of the actor," Varig Airlines, 467 U.S. at 813, 104 S.Ct. at 2764, and use a two-step process to examine the challenged conduct. See Berkovitz v. United States, 486 U.S. 531, 536, 108 S. Ct. 1954, 1958-1959, 100 L. Ed. 2d 531 (1988); In re Joint E. & So. District Asbestos Litigation, 891 F.2d 31, 36 (2d Cir.1989). First, we must consider whether the conduct was a matter of choice for the acting employee, and if so, we must then determine whether the employee's actions involved public policy considerations. Asbestos Litigation, 891 F.2d at 36.

As to the first step, the Supreme Court has determined that "conduct cannot be discretionary unless it involves an element of judgment or choice." Berkovitz, 486 U.S. at 536, 108 S.Ct. at 1958-1959. Thus, when a federal statute, regulation, or policy specifically dictates a course of action for an employee to follow, the employee has no option but to adhere to the directive. Id. If the employee can make no choice, then there can be no discretion for the exception to protect. Id. No such directive was applicable to Dr. Baer in this case; at the critical points in these events he had the power to choose his actions.

But this does not end the inquiry. As we have noted, to qualify for the exception, Dr. Baer's critical choices had to involve policy considerations, and here they did not. The Berkovitz Court stated that "[t]he exception, properly construed, * * * protects only governmental actions and decisions based on considerations of public policy." Id. at 537, 108 S.Ct. at 1959. If an action leaves "room for policy judgment and decision there is discretion." Dalehite, 346 U.S. at 36, 73 S.Ct. at 968.

In Indian Towing Co. v. United States, 350 U.S. 61, 76 S. Ct. 122, 100 L. Ed. 48 (1955), the Court distinguished between an initial government decision to build a lighthouse and subsequent actions to maintain the lighthouse. The initial decision to undertake and maintain a lighthouse service was a discretionary judgment; however, the failure to adequately maintain the lighthouse in good condition, since it did not involve any permissible exercise of policy judgment, subjected the government to suit under the FTCA. Id. at 69, 76 S.Ct. at 126-127; see also Berkovitz, 486 U.S. at 538 n. 3, 108 S. Ct. at 1959 n. 3.

Similarly, in Caban, we determined that the adoption of a regulation by the INS was discretionary but a wrongful detention by an INS employee was not discretionary. Caban, 671 F.2d at 1233. See also Caraballo v. United States, 830 F.2d 19, 21-22 (2d Cir.1987) (initial decision to patrol a national park was discretionary, but the manner in which decision was executed was not discretionary); Eklof Marine Corp. v. United States, 762 F.2d 200, 204-05 (2d Cir.1985) (initial decision to provide navigational aid by marking obstructions was discretionary, but the precise manner of marking an obstruction did not implicate policy and was not a discretionary act).

The government tries to distinguish this line of cases in two ways. First, it contends that the Indian Towing approach is inapplicable because the government here did not decide upon a particular safety measure and then negligently implement its decision. The government would thus restrict Indian Towing to cases involving a decision to provide safety measures. We decline to limit Indian Towing so narrowly. We think that case rested not so much on the government's decision to install a safety device, a lighthouse, but on its negligent maintenance of the lighthouse and the fact that decisions to maintain the lighthouse did not implicate policy considerations. In Indian Towing, only the initial decision to install the lighthouse was discretionary.

Sometimes, even the execution of a policy decision is subject to protection by the discretionary function exception; this occurs when the steps in the execution process also require policy judgments. For example, in Varig Airlines, the Court not only found that the Federal Aviation Administration's implementation of a program for compliance review was discretionary, but it also found that the acts of the FAA employees who executed the program were protected, because those employees "were specifically empowered to make policy judgments regarding the degree of confidence that might reasonably be placed in a given manufacturer, the need to maximize compliance with FAA regulations, and the efficient allocation of agency resources." Varig Airlines, 467 U.S. at 820, 104 S.Ct. at 2767-68. However, in Caban, we determined that even if regulations and statutes appear to vest immigration officials with broad discretion, where the language goes only to the standard of care by which governmental employees' behavior is to be judged, the language does not convert the discharge of prescribed responsibilities into decisions which involve a choice between competing policy considerations. Caban, 671 F.2d at 1233. In short, the key inquiry is whether the decisions, be they initial decisions or implementation decisions, were necessarily susceptible to policy analysis. Asbestos Litigation, 891 F.2d at 37; United States Fidelity & Guar. Co. v. United States, 837 F.2d 116, 121 (3d Cir.), cert. denied, 487 U.S. 1235, 108 S. Ct. 2902, 101 L. Ed. 2d 935 (1988).

In this case, the government's decisions to encourage rabies research, provide funding for the research, and provide NYSDOH with the virus were all decisions that implicated policy considerations and therefore were protected by the discretionary function exception. The same is not true, however, with Dr. Baer's alleged failure to warn of the extreme dangers presented by the particular circumstances of the March 29th experiment which Dr. Baer failed to interrupt. The situation simply did not lend itself to policy balancing, not is there any indication that Dr. Baer considered the policy implications or the pros and cons of allowing the experiment to proceed. Once he became aware of the risks, Dr. Baer was obligated to give warning. His is precisely the type of negligent omission for which congress waived the government's sovereign immunity in order to provide redress to citizens for the torts of government actors.

The government relies on Dalehite to support its contention that even mundane decisions necessary to execute a policy decision are entitled to protection under the discretionary function exception. Dalehite involved a tragic explosion of fertilizer that had been manufactured and shipped at the instigation and under the supervision of the government. The Supreme Court held that even those actions that were necessary in the execution of a policy decision were entitled to protection under the discretionary function exception. Dalehite, 346 U.S. at 36, 73 S.Ct. at 968. Dalehite is distinguishable, however, because the Court could not find there any specific instance of negligence by an employee. The accident was "a complex result of the interacting factors of mass, heat, pressure and composition," id. at 42, 73 S.Ct. at 971, and no single action was identified as the cause of the accident. Liability of the government thus could only have been based on the larger policy decision to manufacture and store the fertilizer.

The government suggests that a similar inability to pinpoint the precise source should prevent liability in this case, because this accident, too, was the result of complex processes. The district court, however, was able to penetrate the complex facts of this case and determine that Jerome Andrulonis's illness resulted, at least in part, from Dr. Baer's failure to warn of the potentially disastrous consequences of the March 29th experiment he was observing. This was a nondiscretionary negligent omission that did not implicate any balancing of policy factors. The district court's findings on this point are not clearly erroneous, and the government is therefore not entitled to protection from suit under the discretionary function exception.

The district court found the government liable on three independent theories of tort liability. We need discuss only the supplier's duty to warn, because it is sufficient to affirm on the liability issue.

Under the FTCA, the law of the state where the tort occurred applies. 28 U.S.C. Sec. 1346(b); Hatahley v. United States, 351 U.S. 173, 182, 76 S. Ct. 745, 752, 100 L. Ed. 1065 (1956); Montellier v. United States, 315 F.2d 180, 185 (2d Cir.1963). Because most of the acts or omissions of Dr. Baer occurred in New York, the substantive law of New York governs this case.

Under established New York law, a supplier of a product which the supplier knows or should know is dangerous if used in the usual and expected manner has a duty to adequately warn users of the product of the danger unless the danger is obvious or well-known. Young v. Elmira Transit Mix, Inc., 52 A.D.2d 202, 204-05, 383 N.Y.S.2d 729, 731 (4th Dept.1976) (Cardamone, J.) (citing Restatement (Second) of Torts Sec. 388); Billiar v. Minnesota Mining and Mfg. Co., 623 F.2d 240, 243 (2d Cir.1980). See also Schumacher v. Richards Shear Co., Inc., 59 N.Y.2d 239, 246-47, 464 N.Y.S.2d 437, 441, 451 N.E.2d 195 (1983). Breach of this duty can lead to liability in negligence.

The government initially argues that a supplier's duty to warn is inapplicable to this case because the doctrine should be limited to a commercial context; that purely scientific activities should be exempt from the rigors of traditional tort law; that the risk-spreading rationale underlying imposition of this duty should not be applicable because the government is not an income-producing entity; and that imposing liability in this situation would tend to curtail scientific research, and restrict the flow of information and exchange of knowledge which are vital for producing health-related benefits.

Although we are sympathetic to the government's policy arguments, they simply are not supported by the applicable law. In the first place, the FTCA specifically provides for the imposition of liability on the federal government as if it were a private person. The fact that the government does not generate income and is not considered a commercial entity is irrelevant in light of congress's decision that governmental immunity is waived in certain circumstances if government employees commit torts.

Furthermore, we agree with the district court that New York does not require a commercial transaction as a prerequisite to liability for breach of a duty to warn. Restatement (Second) of Torts Sec. 405; Pease v. Sinclair Refining Co., 104 F.2d 183, 186 (2d Cir.1939) (applying New York law). In addition, and contrary to the government's contention, there is probably no entity better able to spread the risk of scientific research than the government, which has the biggest "customer" base of all--taxpayers. Finally, in a situation where our entire population may benefit from eradication of rabies, it is not unreasonable to have the potential beneficiaries share the cost of an unfortunate incident occurring in the search for an effective method for vaccination against this lethal disease.

Although the imposition of a duty to warn of potential hazardous consequences may, as the government argues, tend to chill the collegial atmosphere of the scientific community, we think that the benefits of imposing such a duty, at least in the highly unusual circumstances of this case, far outweigh any potentially chilling effects. All scientists should assure that reasonable and appropriate safety precautions are taken in the lab. Laboratory staffs should not have to bear the financial burdens in addition to the personal risks that accompany the quest for scientific advancement. If identifiable negligence does occur, its economic consequences need not be excused as necessary for the benefit of science, no matter how noble the goals of the supervising scientists.

The government also contends that even if the imposition of a supplier's duty to warn is the appropriate legal standard, Dr. Baer satisfied this duty by informing Dr. Debbie of the extremely high concentration of the virus. The district court found that at the time Dr. Baer gave the warning, it was sufficient merely to advise Dr. Debbie that the viral strain was highly concentrated; given what Dr. Baer knew at that time, the warning was proportionate to the risk. However, the district court also found that when Dr. Baer actually observed the experiment, he should have realized the extreme danger created by using his ERA-BHK/21 virus in a leaky aerosolizing machine and should have immediately supplemented his warning. The government argues that requiring Dr. Baer to supplement the warning just because he was a rabies expert and he decided to stay at Griffin to observe the experiment is incorrect.

Under the applicable New York law, we disagree. In Cover v. Cohen, 61 N.Y.2d 261, 274-75, 473 N.Y.S.2d 378, 385, 461 N.E.2d 864 (1984), the New York Court of Appeals determined that a manufacturer or retailer may incur liability for failing to warn of newly discovered dangers in the use of a product that come to his attention after manufacture or sale. These dangers may be discovered through advancements in the state of the art, with which the manufacturer is expected to stay abreast, or through being made aware of later accidents involving dangers in the product, of which warning should be given to users. Although a product may be reasonably safe when manufactured and sold and involve no then-known risks of which warning need be given, risks thereafter revealed by user operation and brought to the attention of the manufacturer or vendor may impose upon either or both an additional duty to warn. Id. at 275, 473 N.Y.S.2d at 385, 461 N.E.2d 864. The type of notice that will trigger a manufacturer's post-delivery duty to warn is a function of the degree of danger involved and the number of instances reported, id., and these are questions of fact. Id. at 276, 473 N.Y.S.2d at 385, 461 N.E.2d 864.

In this case, Judge Munson determined that Dr. Baer, while observing the experiment in the lab, should have known of the dangers involved and that he had sufficient notice to trigger a duty to provide additional warning. Dr. Baer was observing an experiment with the extremely potent rabies virus he had supplied being used in a leaky machine in a way that could potentially cause great harm to those present in the lab. In these circumstances, he should have realized the risks and warned against continuing the experiment without additional precautions.

Judge Munson based his conclusion on several subordinate findings: (1) whenever aerosols were used at the CDC labs in Georgia, a laminar flow containment system was also used; (2) Dr. Baer knew that safe lab practice required containment; (3) Dr. Debbie told Dr. Baer about the red dust that had escaped from the Uni-Glatt machine during the third run of the experiments; (4) Dr. Baer saw nonpareils escape from the machine; (5) Dr. Baer saw Dr. Debbie patch the machine with electrical tape; (6) photographs of the machine showed that it did not appear to be airtight; and (7) Dr. Baer saw that the machine had to be assembled by hand in the laboratory for each experiment. Judge Munson also found that Dr. Baer's testimony, denying his knowledge of the Uni-Glatt's leak, lacked credibility.

When Dr. Baer saw that the Uni-Glatt was not airtight, he should have immediately alerted Dr. Debbie and the lab staff, including Andrulonis, of the hazards posed by the way the ERA-BHK/21 virus was to be used; his failure to do so was, under New York law, a breach of duty to those in the lab, including Andrulonis.

Finally, the government argues that even if the case is properly a duty-to-warn action, and even if Dr. Baer's single preliminary warning was inadequate, the government still cannot be liable, because the case falls within the "knowledgeable user" exception. This argument focuses on proximate cause, which is a necessary predicate to finding liability. Sheehan v. City of New York, 40 N.Y.2d 496, 501, 387 N.Y.S.2d 92, 95, 354 N.E.2d 832 (1976).

In duty to warn cases, New York recognizes two circumstances that would preclude a finding of proximate cause: obviousness and the knowledgeable user. Proximate cause cannot be found when the dangers are obvious or well-known. Lancaster Silo & Block v. Northern Propane Gas, 75 A.D.2d 55, 427 N.Y.S.2d 1009, 1015 (4th Dept.1980). Nor can it be found with a knowledgeable user, one who is actually aware of the dangerous nature of the product supplied. Belling v. Haugh's Pools Ltd., 126 A.D.2d 958, 959, 511 N.Y.S.2d 732, 733 (4th Dept.), appeal denied, 70 N.Y.2d 602, 518 N.Y.S.2d 1024, 512 N.E.2d 550 (1987). Although the government relies here solely on the knowledgeable user doctrine, loosely referred to as the "knowledgeable user exception", a brief look at "obviousness" is necessary to understand why the government's argument fails.

If a danger is obvious, there is no duty to warn against it. Kerr v. Koemm, 557 F. Supp. 283, 287 (S.D.N.Y.1983). The Restatement (Second) of Torts Sec. 388 and comment (k) state that a duty to inform users of a danger exists if the supplier of a product has no reason to expect that the user will discover its condition and realize the danger involved. See also Lancaster Silo, 427 N.Y.S.2d at 1015; Young, 383 N.Y.S.2d at 731; Kerr, 557 F. Supp. at 287. This inquiry into the obviousness of the danger depends not upon actual knowledge of the user, but upon whether the danger was sufficiently obvious that it would be unreasonable to impose a duty to warn on the manufacturer. Kerr, 557 F. Supp. at 287. Thus, the focus of the "obviousness" inquiry is upon the objective reasonableness of the supplier's judgment about whether users will perceive the danger. Id. The danger must be so apparent or so clearly within common knowledge that a user would appreciate the danger to the same extent that a warning would provide. Belling, 511 N.Y.S.2d at 733. The district court held that the danger here was not obvious because Dr. Baer actually witnessed the experiment and should have immediately recognized that neither Dr. Debbie nor Andrulonis realized the hazards they were facing. On appeal the government does not challenge that conclusion.

In contrast, the "knowledgeable user exception", on which the government does rely, involves a subjective test: whether the particular user was aware of the danger. McDaniel v. Williams, 23 A.D.2d 729, 257 N.Y.S.2d 702 (1st Dept.1965); Rosebrock v. General Electric Co., 236 N.Y. 227, 140 N.E. 571 (1923); Kerr, 557 F. Supp. at 287. Its rationale is that knowledge is equivalent to prior notice. Billiar, 623 F.2d at 243. Where the plaintiff knows of a danger, a warning cannot increase his awareness of its presence, Kerr, 557 F. Supp. at 286, and where a warning would not have prevented the harm, a failure to warn cannot be the proximate cause of the injury. Torrogrossa v. Towmotor Co., 44 N.Y.2d 709, 711, 405 N.Y.S.2d 448, 449, 376 N.E.2d 920 (1978); Billiar, 623 F.2d at 243.

In this case, the district court determined that Dr. Debbie did not have actual knowledge of the specific danger in the March 29th experiment. Of course, it determined that Dr. Baer did not have actual knowledge of the specific danger either. Thus, the question becomes whether the knowledgeable user exception applies where the knowledge of both supplier and user is imputed, not actual. The government suggests that the exception should apply here, because the district court found that the state was more than twice as culpable as the federal government, and that Dr. Debbie's "failure to appreciate the hazard to which he was exposing his workers seems more unreasonable than Dr. Baer's failure to recognize the same hazard." Andrulonis, 724 F. Supp. at 1508.

While attractive, the government's argument is not dispositive, because the knowledgeable user exception applies only where the user is actually aware of the precise danger involved. Belling, 511 N.Y.S.2d at 733; Kerr, 557 F. Supp. at 287. A supplier's duty to warn, on the other hand, arises if he knows or should know of the danger; actual knowledge of the precise danger is not required to hold the supplier liable. Schumacher, 59 N.Y.2d at 248, 464 N.Y.S.2d at 441, 451 N.E.2d 195; Young, 383 N.Y.S.2d at 731. Here, the district court determined that the user had no actual knowledge of the precise danger, but both the supplier, Dr. Baer, and the user, Dr. Debbie, were charged with imputed knowledge. In these circumstances, the government, through Dr. Baer's breach of his duty to warn, is liable to Andrulonis for his injuries.

Since we find that the duty-to-warn theory supplies a sufficient basis to establish the government's liability and to affirm that part of the district court's judgment, we need not address the district court's further reliance on the other two theories of liability.

Before the end of trial, the Andrulonises and the non-governmental defendants entered into a series of settlement agreements resulting in a total settlement of $1.3 million. The settling parties agreed to allocate $225,000 to Jerome Andrulonis and $1,075,000 to Joanna Andrulonis. Any settlement funds allocated to Joanna Andrulonis, whose claims against the government were dismissed for failure to timely file an administrative claim, could not be used to reduce the final award of damages to Jerome Andrulonis as provided under section 15-108 of New York's General Obligations Law. The government and NYSDOH now challenge this allocation and the district court's approval of it as being contrary to the policy underlying that section.

Section 15-108 of New York's General Obligations Law provides an incentive to settle and attempts to define the effect of a settlement on collateral rights and liabilities in future litigation. Section 15-108(a) provides in relevant part:

Effect of release of or covenant not to sue tortfeasors. When a release * * * is given to one of two or more persons liable or claimed to be liable in tort for the same injury, or the same wrongful death, it does not discharge any of the other tortfeasors from liability for the injury or wrongful death unless its terms expressly so provide, but it reduces the claim of the releasor against the other tortfeasors to the extent of any amount stipulated by the release or the covenant, or in the amount of the consideration paid for it, or in the amount of the released tortfeasor's equitable share of the damages under article fourteen of the civil practice law and rules, whichever is the greatest.

The "right of set-off" provided by subdivision (a) "is intended to avoid a result in which the non-settling tortfeasors bear more than their equitable share of the plaintiff's damages." Lambert Houses Redevelopment Co. v. HRH Equity Corp., 117 A.D.2d 227, 232-33, 502 N.Y.S.2d 433, 436 (1st Dept.1986). The government and the state contend that the apportionment of settlement proceeds in the case at bar undermines this purpose. They claim that the Andrulonises intentionally allocated most of the settlement proceeds to Joanna Andrulonis in order to minimize any right of setoff that might arise from trial of Jerome Andrulonis's remaining claims. Finally, they argue that the court should not have accepted the settling parties' decision, but should have independently allocated the settlement funds to reflect the relative injuries suffered by Jerome and Joanna Andrulonis.

In Hill v. St. Clare's Hospital, 67 N.Y.2d 72, 77, 499 N.Y.S.2d 904, 907, 490 N.E.2d 823 (1986), the injured party and his wife, as derivative claimant, entered into a settlement with the original tortfeasors, but continued their claim against subsequent tortfeasors for aggravation of the original injuries. Their general release contained no reservation of rights and no allocation of damages as between plaintiffs or as between the successive injuries. Id. Although the court focused primarily on the problem of allocating settlement proceeds between original injuries and aggravated injuries for purposes of setoff against the subsequent tortfeasors, its analysis is helpful in determining a proper apportionment between two plaintiffs.

The court stated that "[t]o the extent that the injured party and the original tort-feasor are permitted to stipulate the amount by which the liability of the successive tort-feasor is to be reduced by the original tort-feasor's payment, intent remains a factor, but the amount stipulated must also be shown to have been arrived at in good faith." Id. at 84-85, 499 N.Y.S.2d at 912, 490 N.E.2d 823. The court determined that the finder of fact, in making a decision as to attribution of the settlement payment, should "consider the statement as to allocation, if any, * * * contained in the settlement documents and the gravity of the respective injuries and determine whether the amount allocated by the parties was arrived at in good faith." Id. at 86, 499 N.Y.S.2d at 912, 490 N.E.2d 823 (emphasis added).

In Casey v. State, 119 A.D.2d 363, 507 N.Y.S.2d 159 (2d Dept.1986), the plaintiffs sued three alleged tortfeasors for wrongful death and for conscious pain and suffering. Before trial, a settlement was reached with two of the defendants, but neither release separately allocated the proceeds to the wrongful death and the pain and suffering causes of action. Instead, the Surrogate's Court made the allocation decision. The court held that any settlement sum, even though allocated by the parties or by a decree of the Surrogate's Court, must, for the purpose of a subsequent setoff, be apportioned at the trial; and it is the obligation of the finder of fact, after hearing the evidence, to make such an apportionment. Id. 507 N.Y.S.2d at 162. A nonsettling party is not bound by any agreement as to allocation made without its knowledge or consent. Id. 507 N.Y.S.2d at 163. Therefore, the self-created apportionment proffered by the claimants and approved by the Surrogate was not binding on the trial court. Id.

In Casey, the court was concerned that the defendant state would be responsible for the bulk of a wrongful death award because the claimants' proposed allocation would have resulted in $175,000 of the settlement being allocated to a "potentially nonexistent pain and suffering cause of action", thereby unfairly reducing the setoff in the wrongful death action and leaving the state with none of the benefits of section 15-108. Id. Thus, the court, per Justice, now District Judge, Arthur Spatt, stated: "We are dealing with two separate and distinct causes of action. The State is entitled to a true setoff based on the merits of the respective causes of action. This apportionment [of the earlier settlement] must be predicated on the evidence adduced at the trial with regard to the monetary value of each cause of action." Id.

Similarly, in Merrill v. State of New York, 110 Misc. 2d 260, 442 N.Y.S.2d 352 (1981), aff'd, 89 A.D.2d 802, 453 N.Y.S.2d 383, 384 (4th Dept.1982), where plaintiffs had earlier settled with one of the joint tortfeasors for a gross unallocated amount in a case involving separate claims for personal injuries and loss of services, the trial court held that the proper way to apportion the settlement proceeds was in proportion to the damages found to have been sustained by the respective claimants. See also Hager v. Hutchins, 91 Misc. 2d 402, 398 N.Y.S.2d 316 (1977) (trial court confronted with a settlement with several defendants for an unallocated sum in a case involving causes of action for wrongful death and conscious pain and suffering held that a prior order of the Supreme Court which had approved the compromise of the claims against the settling defendants and allocated the proceeds entirely to the wrongful death cause of action was not binding on the trial court).

The district court in this case assessed the reasonableness of the apportionment of the settlement monies and ultimately approved the allocation. It rejected the government's suggestion that Hill and Casey required that for setoff purposes, the apportionment of a settlement fund must in every instance approximate the relative injuries of the plaintiffs. Instead, the district court conducted a good faith analysis to determine whether the amounts allocated to Jerome Andrulonis and Joanna Andrulonis could be approved. First, it asked whether the amount set aside for Jerome Andrulonis was a reasonable approximation of the settling defendants' equitable share of his total damages, and second, whether the amount set aside for Joanna Andrulonis was unreasonably excessive in light of the damages she actually suffered.

The district court found that the amount allocated to Jerome Andrulonis did not grossly underestimate the settling parties' equitable share of the damages he suffered. Although Judge Munson noted a significant discrepancy between Lilly/Thompson's settlement amount and its equitable share of Jerome Andrulonis's damages, he considered all the settling defendants as a group and determined that the total amount Jerome Andrulonis received did not so grossly deviate from the amount ultimately allocated to those defendants as to raise an inference of bad faith.

Judge Munson then turned to Joanna Andrulonis's portion of the settlement proceeds. In determining whether the total amount she received was a good faith assessment of the value of her claims, he applied the standard used to review a jury damage award--whether the award was so excessive as to "shock the judicial conscience". See Martell v. Boardwalk Enterprises, Inc., 748 F.2d 740, 750 (2d Cir.1984). The district judge felt this was the appropriate standard because a settlement precluded a right to a jury trial. Sensing no shock and looking to other damage awards in loss of consortium actions, the district judge concluded that the apportionment of settlement monies was made in good faith.

The district judge relied in part on the facts that Joanna Andrulonis was procedurally barred from any recovery against the government or the state, and that any amount she received from settling her claims with the private defendants represented the entirety of the damages she will recover. Thus, the nongovernmental defendants would bear the entire cost of the damages suffered by Joanna Andrulonis, including that portion for which the United States and the NYSDOH would otherwise be responsible, but neither the government nor the state would be required to contribute.

We disagree with that analysis. Judge Munson approved the settling parties' allocation of settlement proceeds after determining whether it was "reasonable." New York law, however, requires the trial court to make an independent determination of what the proper apportionment of settlement proceeds should be, based on the monetary value of each cause of action. See Casey, 507 N.Y.S.2d at 163. Thus, the trial court should have made its allocation decision de novo, and not just reviewed the parties' chosen apportionment for reasonableness.

In addition, while the fact that Joanna Andrulonis had no claim against either the government or the state may have influenced the parties in deciding whether or not to settle, it is not conclusive under New York law of the amount that can be allocated to Joanna Andrulonis upon a subsequent setoff. If Joanna Andrulonis had timely filed an administrative claim she also could have recovered damages from the government, and it may be unfortunate that she will not be able to recover from the governmental tortfeasors; but New York does not permit the parties' private agreements to decrease the amount of the settlement to be offset against the non-settling defendants. The district court's point that only the settling defendants are paying for Joanna Andrulonis's injuries is incorrect, since the government and the state are, through the corresponding reduction in the setoff on Jerome Andrulonis's claim, being required to carry a significant portion of her award.

Settling parties may not structure the apportionment to avoid a later setoff by a nonsettling defendant; to hold otherwise would permit them to circumvent the policy underlying section 15-108. The district court recognized that this is precisely what the Andrulonises had tried to do, but it felt that New York law did not prohibit such a result. We disagree. As shown above, the New York cases establish that the parties' settlement agreement, even when confirmed by a court order, does not bind a trial court in determining the proper allocation of settlement proceeds for purposes of setoff. That allocation must follow from a comparison by the trial court of the injuries of the settling plaintiffs. We therefore remand to the district court with a direction to independently evaluate the relative injuries of the settling plaintiffs and to reduce the damages to be awarded to Jerome Andrulonis by the amount of the settlement thus allocated to him.

D. Apportionment of Culpability between the Government and NYSDOH

The district court, pursuant to N.Y.C.P.L.R. Secs. 1401-1404 governing contribution claims, found the government responsible for 30 percent of the damages suffered by Jerome Andrulonis and NYSDOH responsible for 65 percent. NYSDOH appeals this apportionment and contends that the court erred in finding it responsible for more of the damages than the government. Relying on the notion that culpability is correlated with duty, which in this case happened to depend on the expertise of the actor, NYSDOH argues that the percentages of fault must be reassessed to reflect the federal government's greater responsibility in causing Jerome Andrulonis's illness. NYSDOH argues that the district court's finding that Dr. Baer was the preeminent expert in the field of rabies mandates that the government, having greater expertise, should be charged with a greater share of the damages.

We disagree. NYSDOH is correct in asserting that New York law holds experts to a higher standard of care. See Restatement (Second) of Torts Sec. 289, comment m; Toth v. Community Hospital at Glen Cove, 22 N.Y.2d 255, 263, 292 N.Y.S.2d 440, 447-48, 239 N.E.2d 368 (1968) (holding specialists accountable to a higher standard than general practitioners); Hope v. Fall Brook Coal Co., 3 A.D. 70, 75, 38 N.Y.S. 1040, 1043 (4th Dept.1893) (duty of care is measured against knowledge and expertise of the participants). However, NYSDOH confuses application of the appropriate standard of care in defining the minimal level of duty that may be owed to another, with the entirely different process of weighing the expertise of an actor in determining his portion of damages. In essence, NYSDOH urges that the percentages of damages apportioned must be directly related to the relative levels of expertise of the defendants. New York law does not require such a restrictive approach.

The district court's apportionment of damages, like other findings of fact, is subject to the clearly erroneous standard of review. Fed.R.Civ.P. 52(a). The district court relied on Dr. Baer's expertise in concluding that he had a duty to warn and, thus, in holding the government liable to Jerome Andrulonis. Turning to the degree of the state's culpability for purposes of contribution, the court had before it evidence of Dr. Debbie's expertise with the rabies virus, his prior experience with the Uni-Glatt machine, his intimate knowledge of conditions at the Griffin laboratory, and his role as supervisor of the laboratory with the attendant responsibilities. Balancing all the circumstances of the case, Judge Munson found Dr. Debbie's failure to appreciate the hazards of using the leaky machine in this series of rabies experiments to be more culpable than Dr. Baer's single tragic lapse. Since his findings are not clearly erroneous, we affirm his apportionment of damages under N.Y.C.P.L.R. Sec. 1402.

The district court awarded Jerome Andrulonis $2,417,238 for his future custodial care, relying on an assumption of continued care within the home. Jerome Andrulonis cross-appeals from this part of the judgment, claiming that the damage figure must be based on care in a neurological facility, since it is uncertain whether Joanna can continue to care for Jerome in her home indefinitely into the future. He asks that the damages for future custodial care be increased to $6,841,925.

This argument is without merit. The district court estimated future custodial care on the basis of 24-hour care with three shifts of nursing aides in his home for the rest of his life. Although professional nurses were not considered necessary for Jerome's daily care, the district court used the salary for professional nurses because no one else was willing to do the job at a lower cost. In addition, the district court also determined that Jerome would function best in the home environment where he could live in a predictable structure provided by his family. Since the plaintiff presented no evidence as to when, if ever, institutional care would replace the completely supervised home care presently being provided for Jerome, the district court was not clearly erroneous in the amount it awarded for future custodial care.

We reverse that portion of the judgment of the district court that calculated the setoff under New York's General Obligations Law Sec. 15-108 from the parties's own allocation of settlement proceeds and remand on that issue for further proceedings consistent with this opinion. In all other respects the judgment is affirmed.