Are IAEA Safeguards on Plutonium Bulk-Handling Facilities Effective?

Marvin M. Miller

Department of Nuclear Engineering Massachusetts Institute of Technology

Third in a series of papers on issues bearing on extending and strengthening the Nuclear Non-Proliferation Treaty.

Marvin M. Miller is a senior research scientist with the Department of Nuclear Engineering and the Center for International Studies at MIT. Miller has served as a Foster Fellow with the Nuclear Weapons and Control Bureau of the U.S. Arms Control and Disarmament Agency (ACDA) and is currently a consultant on proliferation issues for ACDA, Los Alamos and Oak Ridge National Laboratories and Nuclear Control Institute.

August 1990

The purpose of this paper is to assess the effectiveness of international (IAEA) safeguards at peaceful nuclear fuel cycle facilities which handle plutonium in bulk form. There are two facilities of this type: reprocessing plants which extract the plutonium from nuclear fuel irradiated in nuclear reactors, and fabrication plants which process the extracted plutonium into fresh fuel assemblies.

I. Introduction

The rationale for the use of plutonium in the nuclear fuel cycle is the contention that such use is the only means of insuring the viability of nuclear power as a long-term energy source. That is, the present generation of nuclear power reactors are too inefficient in the use of uranium to sustain a large contribution of nuclear power, say, 1000 GWe worldwide, beyond the next century. The only way to extend this time horizon significantly is to use plutonium-fueled fast breeder reactors. For this reason research, development and demonstration of breeder reactors and the associated fuel cycle facilities were an early feature of the nuclear power programs in the major industrialized countries.

The vision of nuclear power as an inexpensive and environmentally-benign source of energy has dimmed considerably in recent years mainly because of widespread concerns about reactor safety, as well as the disposal of radioactive wastes, and the possible misuse of peaceful nuclear facilities and materials by both states and sub-national groups for the production of nuclear weapons. In response to this reality and the realization that low-cost uranium resources are abundant, the emphasis of nuclear development and demonstration, particularly in the United States, has shifted from rapid movement to a plutonium breeder economy towards the validation of new reactor concepts which have a higher degree of passive safety than existing reactors.

Despite this, the nuclear establishment in several countries, notably France and Japan, still insist that in the long run the plutonium breeder will be needed. To this end, it is necessary to gain experience in plutonium operations via operation of reprocessing and fuel fabrication plants and use of plutonium fuel in both prototype breeder reactors and existing light water reactors.

The counterargurnent is that, given especially the substantial potential for both increased energy efficiency and wider use of various renewable energy sources, the plutonium breeder will not be needed until well into the 21st century (if ever), even if the threat of greenhouse warming limits the use of fossil fuels. From this perspective, the uncertain economic benefits and waste disposal advantages of plutonium reprocessing and recycle in light water reactors are outweighed by the substantial risks of diversion of this material, as well as its potential release to the environment in normal fuel cycle operations, and under accident conditions.

Although the issue of the need for nuclear power in general and plutonium breeders in particular requires the consideration of its comparative economics and environmental impacts as well as its proliferation and terrorism risks, we consider only this last aspect here, and further restrict ourselves to the question of the effectiveness of safeguards at plutonium bulk handling facilities. We begin in the next two sections with a brief sketch of the relevant safeguards background.1

II. Safeguards Goals

In contrast to IAEA document INFCIRC 66/REV. 2, which delineates the safeguards system for nuclear facilities in non-NPT states, the corresponding document that was developed to detail the safeguards obligations of states party to the NPT, INFCIRC/153, provides a technical definition of the safeguards objective, namely "the timely detection of the diversion of significant quantities of nuclear materials from peaceful activities...and deterrence of such diversion by the risk of early detection." The key terms of this objective were not defined in INFCIRC/153; this task was given to the Standing Advisory Group on Safeguards Implementation (SAGSI) of the IAEA, an advisory group of technical safeguards experts.

SAGSI considered the problem of quantifying the safeguards objective for several years. It identified four terms appearing either explicitly or implicitly in the statement of the objective just quoted as in need of quantitative expression. These were: significant quantities, timely detection, risk of detection, and the probability of raising a false alarm. It defined the associated numerical parameters (significant quantity, detection time, detection probability, and false alarm probability) as detection goals.

In 1977, SAGSI submitted numerical estimates for these goals to the Director of Safeguards of the IAEA. A significant quantity (SQ) was defined as "the approximate quantity of nuclear material in respect of which, taking into account any conversion process involved, the possibility of manufacturing a nuclear explosive device cannot be excluded." For plutonium the significant quantity was taken to be 8 kg; for highly enriched uranium (HEU), 25 kg of contained U-235; for low-enriched uranium (LEU), 75 kg of contained U-235.

Detection time (the maximum time that should elapse between a diversion and its detection) should be of the same order of magnitude as conversion time, defined as the time required to convert different forms of nuclear material to the components of a nuclear explosive device. For metallic Pu and HEU, conversion time was estimated as 7-10 days; for pure unirradiated compounds of these materials such as oxides or nitrates, or for mixtures, 1-3 weeks; for Pu or HEU in irradiated fuel, 1-3 months; and for low-enriched uranium, 1 year.

On the basis of common statistical practice, SAGSI recommended a detection probability of 90-95%, and a false- alarm probability of less than 5%.

The values recommended by SAGSI for the detection goals were carefully described as provisional guidelines for inspection planning and for the evaluation of safeguards implementation, not as requirements, and were so accepted by the Agency. However, the view of a sector of the non- proliferation community, which was particularly influential in the U.S. during the Carter Administration, was that unless these goals could be met in practice, safeguards were not effective, and the associated activity, e.g., reprocessing of spent reactor fuel to extract plutonium, posed too great a proliferation risk. This perspective was embodied in the major piece of non-proliferation legislation enacted into law in the U.S. during the Carter Administration, the Nuclear Non-Proliferation Act (NNPA) of 1978. This is particularly true with regard to the importance given to the concept of "timely warning" in the NNPA. "Timely" is taken to be detection of a diversion quickly enough to take diplomatic action to prevent the fabrication and insertion of the diverted material into a first bomb that is otherwise complete. Thus, detection time must be even shorter than conversion time, in order to allow for evaluation and response, e.g., even shorter than 1-3 weeks for Pu or HEU compounds in unirradiated form.

In the author's opinion, the view that the detection goals should be operational criteria for safeguards effectiveness has a certain logic. In particular, a diversion of less than a significant quantity would not provide enough material for a nuclear explosive, and with regard to timely warning, it would obviously be advantageous to know about a diversion in time to do something about it, that is, before the diverter could assemble a weapon from the diverted material. However, in judging safeguards effectiveness, it is also logical to be able to distinguish between a situation in which the performance of the safeguards system is only marginally worse than the detection goals, and one in which the differences are considerable. A relevant example would be two reprocessing plants in which the minimum amount of diverted plutonium which could be detected with high confidence is 10 kg/yr and 250 kg/yr, respectively. This raises the issue of what other factors besides numerical goals might be taken into account in judging safeguards effectiveness. Before taking up this issue, we first assess the current situation with regard to the capability of the safeguards system in meeting the detection goals, and the prospects for future improvements.

III. Material Accountancy

The paragraph in INFCIRC/153 immediately following the one in which the safeguards objective is defined delineates the methods to be used in timely detection of diversion:

To this end the Agreement [between the Agency and the State] should provide for the use of materials accountancy as a safeguards measure of fundamental importance, with containlnent and surveillance as important complementary measures.
Application of material accountancy by the IAEA to the detection of diversion of nuclear material is analogous to a bank examiner's financial audit.2 First the nuclear facility operator [bank management] must prepare a material balance [financial statement] covering a specified period, e.g., one year, showing that all nuclear material [money] can be accounted for. More specifically, adding the material inputs (I) [credits] and subtracting the removals (R) [debits] from the beginning inventory (BI) [opening balance] gives the amount that should be in the ending inventory (EI) [final balance]. The IAEA inspector [bank examiner] performs an independent check on at least some of the data presented by the facility operator [bank management] to confirm the absence of deliberate falsification.

This procedure works well in the context of the financial audit and at nuclear facilities where the nuclear material is present only in the form of identifiable and countable items, e.g., fuel assemblies at power reactors. That is, if the "books do not balance," it is a clear indication that there has been an unrecorded removal of nuclear material [money] from the facility [bank]. In the parlance of nuclear materials accountancy, a positive value of "materials unaccounted for" or MUF, defined by

MUF equals (BI + I - R - EI), (1)

where R includes both product and any lost material indicates a diversion, whereas if MUF = 0 and the operator's data has been authenticated by the inspector, then it is possible to unambiguously conclude that no diversion has occurred.

When eq. (1) is applied to materials in bulk form, a problem arises because it is no longer possible to know any of the terms in the equation exactly. Unlike fuel assemblies, the quantity of bulk materials, such as plutonium in reprocessing and fuel fabrication plants and uranium in fabrication and enrichment plants, can only be measured approximately. As a result, even in the absence of diversion, non- zero values of MUF will be measured, and materials accountancy must rely on statistical tests to distinguish positive values of MUF due to diversion from those due to a chance combination of measurement errors. Whether this is a significant problem in terms of meeting the IAEA's detection goal of detecting an SQ depends on the magnitude of the associated errors-technically specified by the variance of MUF, sigma(MUF)---compared with an SQ.3 Intuitively, if the "noise" of the measurement process, as specified by sigma(MUF), is small compared to an SQ, then diversions of material on the order of an SQ should be detectable with both high confidence and only a small probability of a false alarm. Conversely, if sigma(MUF) greatly exceeds an SQ, then the minimum diversion which can be detected with high confidence and a small false alarm probability will also be much greater than an SQ.

Unfortunately, even if sigma(MUF) is small as a percentage of the quantity of material measured, e.g. less than or equal to 1%, in a plant processing large quantities of material, the absolute value of sigma(MUF) will, over a sufficiently long period of time, exceed an SQ.

A relevant example is the planned 800 tonne/yr Rokkasho reprocessing facility at Aomori in Japan. Assuming that: (1) the plant processes spent fuel with an average total plutonium content of 0.9%; (2) the error in measuring the MUF, specified by sigma(MUF), is dominated by the error in measuring the plutonium input, and is equal to 1% of this input, and (3) the material balance calculation is done once a year, then the absolute value of sigma(MUF) = 72 kg of Pu/yr. It is straightforward to show that the minimum amount of diverted plutonium which could be distinguished from this measurement "noise" with detection and false alarm probabilities of 95% and 5%, respectively, is 3.3 sigma(MUF), or 246 kg in this example, equivalent to more than 30 significant quantities.

Besides the fact that the minimum detectable diversion in such a plant greatly exceeds an SQ, the detection time will also exceed the timeliness goals for the various forms of plutonium in the plant, e.g., the 1-3 week goal for unirradiated plutonium compounds such as the plutonium nitrate product of the plant. There are three reasons for this.

In the first place, while the material balance is measured on a yearly basis, the diversion of 8 kg or more of plutonium could take place at any time following plant startup. (Note that an 800 tonne/yr reprocessing plant might process 4 tonnes of fuel containing 36 kg of plutonium per day for 200 days, with the remainder of the time reserved for scheduled and unscheduled maintenance.) Secondly, the determination of the concentration of plutonium in the input and output accountability tanks, as well as in process tanks, currently requires that tank samples taken by the plant operator and given to the IAEA inspector be shipped back to the IAEA analytical laboratory outside Vienna for measurement. Because of stringent national regulations on the shipment of plutonium, this is often a time-consuming process: delays in measuring samples on the order of months are not unusual.

Finally, in the Agencts view, a false accusation of diversion would be extremely serious, and could discredit the safeguards system. Thus, detecting a diversion means, first, detecting a suspicious event, technically an "anomaly," indicative of a possible diversion, such as a large MUF or a film picture indicating unreported movement of nuclear material. The Agency then attempts to systematically eliminate all other possible explanations, such as larger than estimated measurement errors, unreported material losses, defective safeguards equipment, etc. This process is apt to be very time-consuming, especially if remeasurement is required, and the greater the degree of certainty that is required, the longer the process will take. Thus, detection in the spirit of the timely warning philosophy cannot in practice be realized both because of the "untimely" nature of the measurement process and also because of the Agencts philosophy of being extremely careful to avoid an unjustified accusation of diversion.

One obvious technical fix is to perform material balance measurements more frequently, e.g., weekly. Assuming that the percentage error remains the same, a shorter measurement period implies a smaller absolute value of sigma(MUF) since the plant throughput is smaller. This increases the potential for both greater detection sensitivity and timeliness in the detection of an abrupt diversion. Unfortunately, making inventory measurements in large plants, particularly reprocessing plants, is time consuming and expensive because it involves a shutdown of the plant and a washout of the process equipment. For this reason, only one or two inventory takings per year would be acceptable to the plant operator.

Recognizing these difficulties, the Agency has tried to remedy them in two ways: (1) by considering the implementation of advanced technical approaches, in particular, near-real-time accountancy (NRTA), wherein material balance measurements are made frequently without shutting down the plant, and greater reliance on containment/ surveillance (C/S) measures; and (2) by setting up two other levels of safeguards goals, the inspection goals, which are supposed to reflect actual conditions at the facility, requirements prescribed by the safeguards agreements, and capabilities of safeguards measures, and the accountancy verification goal, defined as the minimum quantity of diverted material which could be detected with 95% and 5% detection and false alarm probabilities, respectively, based on achievable measurement uncertainties; currently, the latter is specified by the Agency as a sigma(MUF) of 1% of the input, e.g., 246 kg of plutonium per year at an 800 tonne/yr plant.

The potential of NRTA and greater reliance on C/S measures for improving the effectiveness of safeguards at bulk-handling facilities is discussed in the next section. As to the remedy of defining alternative safeguards goals, this is widely-and, in the author's view, correctly—perceived as a retreat from the original detection goals, indicative of both the Agency's inability to meet these goals and its unwillingness to admit this fact. The result has been a loss of confidence in IAEA safeguards, particularly in the U.S.

In sum, the IAEA's safeguards detection goals cannot be met at large reprocessing and plutonium fuel fabrication facilities using conventional materials accountancy. Although we have followed the common practice of focusing on safeguards in the chemical process area of a reprocessing plant, it is also important to note that there are significant errors in the current measurements of other plutonium- bearing streams in such a plant. In particular, the fuel hulls and the undissolved plutonium which is filtered out of the process stream before it reaches the separation stages both contain on the order of 0.5% of the plutonium input, respec tively, or 40 kg/yr in an 800 tonne/yr facility. The accuracy with which these streams are measured currendy is on the order of 50% of their magnitudes, or about 20 kg of plutonium. Thus even if the potential of NRTA is realized in the process area, the goal of detecting an SQ will not be satisfied unless significant improvements in these measurements are also realized. We return to this point in the next section.

IV. The Technical Potential for Improved Safeguards Performance

In this section, we briefly discuss the technical potential for higher detection sensitivity and greater detection timeliness in large plutonium buIk-handling facilities, particularly reprocessing plants. We begin with a discussion of Near- Real-Time Accountancy (NRTA), that is, making measurements of the material balance more frequently than in conventional materials accounting, e.g., weekly instead of yearly.

What makes NRTA practical is the feasibility of making frequent measurements of the plant's plutonium inventory without shutting it down. This is accomplished by actual measurement of plutonium in most of the process equipment, and reliance on estimates of the plutonium content of those vessels which are inaccessible to measurement.

As previously noted, use of NRTA implies a smaller absolute value of sigma(MUF) since the plant throughput during the material balance period is proportionately smaller. Thus, for the 800 tonne/yr equals 4 tonne/day reprocessing plant we have been using as an example, the weekly throughput of plutonium would be 252 kg. At a measurement error sigma(MUF) = 1%, the minimum diversion which could be detected with detection and false alarm probabilities of 95% and 5%, respectively, is then about 8 kg.

Weekly material balance takings would also increase the timeliness of diversion detection if the measurement samples could be analyzed more quickly. That is, instead of shipping the samples to the lAEA's analytical laboratory outside Vienna, the measurement techniques employed should be amenable to rapid analysis by the IAEA inspectors on-site. Several techniques of this type are in the development and demonstration stage, and their prospects are promising.

The efficiency of NRTA in detecting a protracted diversion of plutonium, i.e., diversion of a small amount of plutonium per week, whose cumulative total over many weeks exceeds 8 kg, is not as clear. (Note that conventional materials accountancy does not distinguish between abrupt and protracted diversion since measurements are only made yearly). The argument of NRTA proponents can be summarized as follows. Assume that one has a significant data bank of MUF values for a period during which there was no diversion. Then the deviation of MUF values from zero during this period must be due to measurement error.4 Thus, in testing another sequence of MUF values for protracted diversion, one can subtract an estimate of the measurement error derived from the diversion-free data from the sequence of MUF values under test. In this manner, one should be able to effectively decrease the magnitude of sigma(MUF), and hence increase the diversion detection sensitivity.

This argument is intuitively appealing, but not entirely convincing, at least not to the author. While there is a considerable literature which seeks to demonstrate the efficacy of various sequential testing procedures in detecting protracted diversion, their basic assumption is the existence of diversion-free MUF data which can be used as a calibration standard. If diversion begins when the plant starts operation and continues as long as safeguards are applied, then there is no such data, and Avenhaus and Jaech have shown that there is no gain in detection sensitivity using NRTA compared with conventional materials accountancy.5 Moreover, the sequential tests do not give the operator- diverter credit for diversion strategies which are more sophisticated than simply removing a fixed amount of plutonium during successive material balance periods. For example, the operator could also put material into the system during some of the material balance periods. This makes diversion look more like measurement noise, and hence makes it more difficult to detect. It may turn out that a selected menu of different sequential tests may be able to detect all credible protracted diversion scenarios with a detection sensitivity significantly greater than that available from conventional materials accountancy, but this remains to be demonstrated.

Finally, it must be recognized that implementation of NRTA would be labor intensive for both the plant operator and the IAEA, and would provide the latter with a degree of insight into plant operations which, while beneficial from the viewpoint of safeguards effectiveness, might also conflict with the operator's desire to protect proprietary information. Thus, higher safeguards costs as well as some degree of opposition from plant operators are likely.

Regarding the prospects for achieving greater safeguards effectiveness by increased reliance on containment and surveillance (C/S) measures, it should be noted that the IAEA safeguards staff and the safeguards support programs in the member states, particularly the U.S., have invested much time and effort over the past 15 years to develop and implement reliable and effective C/S devices, particularly seals and cameras. There is no doubt that such devices have the potential for fulfilling the role of C/S as an important complement to materials accountancy envisaged for it in INFCIRC/153. For example, the use of cameras to provide surveillance of both the spent fuel pool at a reprocessing plant and the transfer of such fuel to the chop-leach cell can detect attempts to process undisclosed spent fuel in the plant. Another relevant example is the use of seals on the tanks containing the plutonium nitrate product of such a plant to detect unauthorized withdrawals of material.

The problem has always been that, while it should be possible to take credit for C/S measures in judging safeguards effectiveness, no one has figured out a logical way of quantifying this benefit and combining it with the assurance of non-diversion provided by materials accountancy into a combined measure of safeguards effectiveness. Nevertheless, the benefit is real and justifies continued effort to improve the reliability and effectiveness of C/S devices.

The principal caveat to the above is that C/S measures cannot substitute for NRTA, particularly in the process area of a reprocessing plant where it is impractical to monitor the myriad of pipes, valves, pumps and tanks using such devices.

Finally, more accurate measurements of the plutonium content of waste streams, particularly the hulls and sludges in reprocessing plants, are required to approach the 8 kg/yr detection goal. Various non-destructive assay (Nl)A) techniques have been used to measure such streams, but the associated errors are large. Active neutron interrogation shows the most promise, but it is also the most difficult to implement.

V. Conclusion

Assuming, optimistically, that: (1 ) a better measurement of the plutonium in the input accountability tank of a reprocessing plant makes it practical to achieve a sigma(MUF) = 0.5% overall in the chemical process area; (2) the use of NRTA on a weekly basis makes it possible to improve on the detection sensitivity of conventional materials accountancy for protracted diversion by a factor of 4, and (3) the plutonium in hulls and sludges can be measured to 10%, then the minimum detectable abrupt and protracted diversions would be about 5 kg and 35 kg, respectively, in an 800 tonne/yr reprocessing plant.

The situation could be somewhat better in a large mixed-oxide (MOX) fuel fabrication plant because the plutonium streams are easier to measure. For example, the plutonium throughput of a 100 tonne/yr MOX facility which fabricates fuel with an average plutonium concentration of 3% would be 3000 kg. Assuming a sigma(MUF) = 0.3% overall, and monthly NRTA with an "effectiveness gain" of a factor of 2 for protracted diversion compared to conventional materials accountancy, the minimum detectable abrupt and protracted diversion would be about 3 kg and 15 kg, respectively.

Is the considerable effort that will be required to achieve these results worth it? The proposition that it is not is usually based on the argument that states which voluntarily accept safeguards by joining the NPT are not likely candidates for diversion. Thus, even if the safeguards goals cannot be rigorously achieved, or even reasonably approximated, the proliferation risks are minimal.

The usual counterarguments are that: (1) such political "credit" cannot be given to non-NPT states who do not accept safeguards voluntarily, but only as a condition for technology transfer; (2) not all NPT states have impeccable non-proliferation credentials, and in practice, it is difficult to implement a safeguards system based on subjective, and changing, judgments about proliferation risk. In the author's view, these counterarguments have merit. However, the most telling argument for more effective international safeguards is its likely impact on domestic safeguards and the risk of sub-national diversion of plutonium. That is, international safeguards are based on national systems of materials accountancy and control. Without the incentive for greater safeguards effectiveness provided by stringent IAEA safeguards criteria, there might be a tendency to relax domestic safeguards. This, in turn, would increase the risk of suSnational diversion, particularly with the collusion of an insider, who was familiar with the weak points of the safeguards system. A historical case in point is that of the plutonium weapons facilities in the U.S. The "mind-set" of the U.S. Atomic Energy Commission, under whose auspices these plants were built and operated, was that the first priority was maximum production, and that the insider threat was not credible. As a result, materials accountancy and control left much to be desired, and considerable material was unaccounted for. While all of this might have plated out in the process equipment or been released to the environment rather than being diverted, operation of plutonium-handling facilities in a manner which disregards both the insider threat and environmental hazards is unacceptable.

In sum, technical measures, especially NRTA, but also, more reliable and effective C/S, greater at-plant IAEA measurement capability, and more accurate measurements of the plutonium in waste streams, could lead to a significant improvement in the effectiveness of international safeguards at large plutonium-handling facilities. Implementation of such measures would increase public confidence in the ability of the IAEA to minimize the risks of the use of plutonium in nuclear fuel cycles. Until these measures can be implemented and demonstrated, it would be prudent to limit plutonium use to research, development, and demonstration projects. For even if these improvements can be practically achieved, there are still diversion risks as well as environmental hazards associated with large-scale transport of plutonium between reprocessing/fabrication plants and reactors. With the current glut of low-cost uranium, the world can afford to take the time to investigate the feasibility of reactors, including breeders, which are not only safer and make waste disposal more tractable, but also have a higher degree of proliferation and terrorism "resistance" than the standard breeder and its associated fuel cycle. If nuclear power is to have a future, it should be in this direction.


1 . The development in Sections 11 and 111 follows that of E.V. Weinstock and J.M. de Montmollin, "IAEA Safeguards: Perceptions, Goals and Performance," Workshop on International Safeguards, Cornell University, Ithaca, New York, May 1984.Back to document

2. This analogy has been used by other authors. See, in particular, J. Lovett, "International Safeguards for Reprocessing CAN Be Effective," Physics and Society, Vol. 19, No. 3, July 1990, pp. 7-9.Back to document

3. To be precise, the error variance of interest is the IAEA inspectors best estimate of MUF, sigma(MUF-D), rather than the variance based on the plant operators measurements alone, sigma(MUF). The difference between these two error variances is not important for the argument here.Back to document

4. In practice, unreported plant losses and unmeasured plant inventory could also lead to non-zero values of MUF. Thus, the effect of these error sources must be differentiated from those due to measurement error before the observed non-zero MUF can be unambiguously identified with measurement error.Back to document

5. R. Avenhaus and J. Jaech, "On Subdividing Matenal Balances in Time and/or Space," Journal of the Institute of Nuclear Materials Managernent, Vol. X, No. 3, 1981, pp. 24-33.Back to document

Return to Nuclear Terrorism Page NCI Home Page