Statement of

GARY MILHOLLIN
Professor, University of Wisconsin Law School and Director,
Wisconsin Project on Nuclear Arms Control

I am pleased to appear before this distinguished Subcommittee to discuss the sale of American supercomputers to foreign entities that develop nuclear weapons. I am a member of the University of Wisconsin law faculty, and I direct a research project here in Washington that is devoted to tracking and inhibiting the spread of nuclear weapons to additional countries.

The Subcommittee has asked me to describe the sales that have happened recently, and to assess their impact on U.S. national security.

The sales and their impact

This past January, Viktor Mikhailov, Russia's Minister of Atomic Energy, shocked the United States government by announcing that his ministry had managed to buy powerful American supercomputers for Russia's nuclear weapon laboratories.

It turned out that Silicon Graphics, Inc., a computer firm headquartered in Mountain View, California, had shipped supercomputers to Chelyabinsk-70, the second most famous nuclear weapon laboratory in Russia, without obtaining the required U.S. export licenses. Chelyabinsk claims to have developed the world's most powerful hydrogen bomb and is roughly equivalent to our Lawrence Livermore National Laboratory. The computers were delivered in the autumn of 1996, at virtually the same time that the White House was turning down requests from IBM and Hewlett Packard to sell computers of equal power to Chelyabinsk.

The White House decision came after a long interagency debate. Virtually all of the concerned agencies--the Departments of Energy, State and Defense--opposed the exports by IBM and Hewlett Packard. So it was a considerable shock to have such a broad decision on U.S. policy overturned by the act of a single exporter, especially one who didn't comply with the law.

Silicon Graphics sold four machines to the Russians, all from its "Power Challenge Deskside" product line. Two were configured with eight microprocessors each and the other two with four microprocessors. By adding additional processors, which are not controlled for export, each computer could be made to operate at 4.4 billion operations per second. According to a Silicon Graphics vice-president, Silicon Graphics also shipped upgrades to the computers in January. Thus, each machine was at least four times more powerful than anything the Russians had before, and according to Mr. Mikhailov, ten times more powerful.

With them, Russia will be able to design nuclear warheads cheaper and faster through simulations and will be able to design more accurate long-range missiles. Mr. Mikhailov has declared to the press that Moscow is still designing new nuclear weapons. Russia will obey the new test ban treaty, he said, but will now design its warheads with simulated explosions--using computers from Silicon Graphics. "Like the United States, we have great expertise in this area," he boasted. In effect, Russia will continue the nuclear arms race on computers made in America.

Russia could also use the machines to do encryption, or to design advanced conventional weapons. Because the machines were shipped without an export license, and are located at a site that is mostly closed to the outside world, Russia can put them to any use it wants.

Only an innocent mistake?

Silicon Graphics claims that it only made an innocent mistake and that it had no idea what Chelyabinsk was up to. Anybody who believes that believes in fairy tales.

Chelyabinsk-70 has been designing nuclear weapons for forty years. In 1992 it was officially declared a Federal Nuclear Center by President Boris Yeltsin. In May 1995, the Commerce Department's Bureau of Export Administration published The Russian Defense Business Directory, a guide to acquaint American exporters with Russia's military sites. The guide listed Chelyabinsk-70's "product line" as the "development of nuclear weapons." Its civilian line was listed as "N\A." The guide stated that Chelyabinsk-70 "has...expertise in the entire development of nuclear weapons, including nuclear physics, hydrodynamics, [and] mathematical modeling...." This was the clearest notice possible to exporters that Chelyabinsk was a nuclear weapon design site. In a memorandum dated January 15, 1997, which Silicon Graphics sent to the Commerce Department, Silicon Graphics admitted that it sold the computers to the "All-Russian Scientific Research Institute for Technical Physics (VNIITF)," which is the official name for Chelyabinsk-70. The memo proves without a doubt that Silicon Graphics knew where the computers were going at the time of sale. Silicon Graphics personnel have admitted to me, in telephone conversations, that Silicon Graphics does a lot of business in Russia. Silicon Graphics has a sales office in Moscow with Russian personnel. It is simply incredible that these Russians would not know the market for supercomputing in Russia, and not know what Chelyabinsk had been doing for forty years.

It is also incredible that Silicon Graphics did not notice that its biggest competitors, Hewlett Packard and IBM, had applied for export licenses to sell to the same buyer. IBM applied after reading about Hewlett Packard's application in the New York Times. But according to Silicon Graphics, not only do its employees not read the Commerce Department's guides to foreign markets, they don't even read about their competitors in the newspapers.

The deal has all the earmarks of a deliberate violation of the law. But whether it was deliberate or not, it was clearly illegal.

Under the U.S. Export Administration Regulations, an American company needs an export license to ship any computer operating above two billion operations per second to a "tier three" country. These countries include China, India, Israel, Pakistan and Russia. There is an exception, which Silicon Graphics has claimed in this case, called the "CTP exception." It provides that a computer that performs between two and seven billion operations per second can be sold without a license to an end-user that is not a nuclear, chemical/biological, missile or military site. Chelyabinsk-70 is both a nuclear site and a military site, so the exception fails on two counts.

If an exporter claims the CTP exception, the exporter must be sure that the end-user qualifies for it. The burden is on the exporter to find out enough about the buyer to determine whether the exception can be used. Silicon Graphics admits that it didn't find out. Its defense is that it didn't ask enough questions. But that is no defense when you have an obligation to ask enough questions. Silicon Graphics either knew that Chelyabinsk was a nuclear and military site, or it didn't bother to find out. Either way, Silicon Graphics broke the law.

I have discussed this view of the export control regulations with career-level experts at the Commerce, Defense and Energy Departments. They all agree that it is correct, and that it is the basis upon which they presently administer export controls.

I urge this Subcommittee to ask the Commerce and Energy Department representatives who are here today to affirm that the burden was on Silicon Graphics to be sure the exception applied. In addition, the Subcommittee should ask them to affirm that Chelyabinsk-70 is a nuclear and military site, and that it does not qualify for the exception.

Helping China too

To make matters worse, Silicon Graphics has acknowledged selling an even more powerful supercomputer to the Chinese Academy of Sciences, which helps develop China's long-range missiles.

Joseph Dinucci, head of corporate marketing for Silicon Graphics, told me on February 21 that his company had shipped the computer to China last spring, also without an export license. The computer sold to China was about twice as powerful as the ones sold to Russia. It performs approximately six billion operations per second. Under Commerce Department regulations, computers performing more than two billion operations per second cannot be shipped to nuclear, chemical/biological, missile or military sites in Russia or China without a Commerce Department export license.

Mr. Dinucci said that Silicon Graphics was "very comfortable" with the sale, which he said was "well executed" and did not require an export license. Mr. Dinucci did not say why the sale did not require a license, but the only exception to the license requirement is for buyers that do not conduct nuclear, chemical/biological, missile or military activities. But according to Chinese government publications, the Chinese Academy of Sciences oversees institutes that perform missile and military research as well as research related to nuclear weapons.

In the 1970s, the Academy helped develop the flight computer for the DF-5 intercontinental missile, which can target U.S. cities with nuclear warheads. The Academy's Mechanics Institute has also developed advanced rocket propellant, developed hydrogen- and oxygen-fueled rockets, and helped develop the shield for the warhead of China's first ICBM. Its Shanghai Institute of Silicate successfully developed the carbon/quartz material used to shield the tip of the reentry vehicle from the heat created by the earth's atmosphere.

The Academy's Institute of Electronics has built synthetic aperture radar useful in military mapping and surveillance, and its Acoustic Institute has developed a guidance system for the Yu-3 torpedo, together with sonar for nuclear and conventional submarines.

In the nuclear field, the Academy has developed separation membranes to enrich uranium by gaseous diffusion, and its Institute of Mechanics has studied the effects of underground nuclear weapon tests and ways to protect against nuclear explosions. It has also studied the stability of plasma in controlled nuclear fusion. Its Institute of Electronics has developed various kinds of lasers used in atomic isotope separation.

According to information published by Silicon Graphics, its "Power Challenge XL" model was sold to the Academy with sixteen processors and is now the "most powerful SMP supercomputer in China." According to information I have received from industry sources, the most powerful computers previously sold to China operated at approximately 1.5 billion operations per second. If this information is accurate, the Silicon Graphics machine is roughly four times more powerful than anything China had before. The new computer, which was financed by a loan from the World Bank, has become the centerpiece of the Academy's new Computer Network Information Center, where, according to Silicon Graphics, it provides China "computational power previously unknown." According to information published by the Academy, the computer is now available to "all the major scientific and technological institutes across China."

This means that any Chinese organization that is designing nuclear weapons or long-range missiles has access to it. In effect, Chinese weapon designers can use the Silicon Graphics machines to design lighter nuclear warheads to fit on longer-range and more accurate missiles capable of reaching U.S. cities. This is a giant loss for U.S. security.

Silicon Graphics has also announced plans to expand the Academy's supercomputer to a full 36 processor configuration, which would more than double its existing power and allow it to operate at above thirteen billion operations per second. The Subcommittee should ask the government today whether that expansion has taken place, and if so, whether it was licensed, and if not, whether it will be licensed in the future.

A disaster waiting to happen

The Commerce Department shares the blame for allowing these sales to happen. Early last year, the administration's nuclear experts asked the Commerce Department to send American computer makers a list of the sensitive nuclear sites in Russia and China. The experts wanted to put the companies on notice, so they wouldn't unwittingly sell high-power machines to these places. The experts wanted to head off exactly the kind of claim that Silicon Graphics is now making. But Commerce refused to publish a list, saying it was against U.S. policy to name such sites in friendly countries.

Why was there a danger of unwitting sales? Because, to please exporters and especially Silicon Valley, Commerce has abruptly slashed export controls on strategic technology to a tenth of what they were in 1992. In 1992, no computer performing more than 12.5 million operations per second could go to Russia or China without an export license. Now, computers up to seven billion operations per second can go without a license, providing the sale is not to a nuclear, chemical/biological, missile or military site.

This has resulted in what the Commerce Department calls an "honor" system. The exporter can ship a powerful supercomputer to a proliferant country without telling anybody, as long as the exporter decides that the computer is going to a safe location. In fact, this is really a "dishonor" system, in which the exporter makes more money if it closes its eyes and holds its nose.

The best proof that this system doesn't work is what Silicon Graphics has done. It has outfitted nuclear and military sites in Russia and China, and has probably done the same in India and Pakistan, all under this "honor" system, and all without telling anybody. If Mr. Mikhailov hadn't bragged to the press about getting his supercomputers, we probably wouldn't know about these sales even today.

The Subcommittee should ask Silicon Graphics to make public all of its sales to China, to Russia, and to all other tier three countries during the past two years. The data should cover all computers performing above two billion operations per second. With this information, the Subcommittee and the public can evaluate how this "honor" system works. It can also evaluate how the sales have affected U.S. national security.

What should happen next?

I have five recommendations.

First, the Subcommittee should urge the Commerce Department to cut off the kind of sales that Silicon Graphics has just made. The "honor system" is broken. It needs to be fixed. To fix it, the Commerce Department should immediately notify U.S. computer makers that they must apply for an export license for any computer operating at more than two billion operations per second shipped to a tier three country. This would stop companies like Silicon Graphics from exporting to known bomb makers while claiming not to know what the bomb makers are doing. If the Commerce Department thinks this change would be an administrative burden, it should furnish data on how many inquiries would be likely to result. Because the Commerce Department is only processing one tenth as many cases now as it did before the end of the cold war, there should still be sufficient staff to do the job.

Second, the Subcommittee should ask the Commerce Department not to approve any additional computer sales to tier three countries without first forwarding the applications to this Subcommittee for oversight review. The Subcommittee should be given fifteen working days to review the applications. Mr. Mikhailov has announced that in 1997, he expects to buy even more powerful American supercomputers, ones that will perform 100 billion operations per second. And Silicon Graphics has announced that it will upgrade the supercomputer at the Chinese Academy of Sciences. The Subcommittee should oversee the Commerce Department's review of such sales.

Third, the Subcommittee should require the Commerce Department to keep the Subcommittee informed of the progress of the Commerce Department's investigation and of its deliberations on penalties. Given the grave consequences of the Silicon Graphics sales, the Commerce Department should consider suspending Silicon Graphics export privileges for six months to a year. That would get industry's attention, and show that our export control laws still mean something. The Subcommittee should also consider barring Silicon Graphics from U.S. defense contracts.

Fourth, the Congress should oppose any further cuts in export controls on computers. The defects of the present system have been created by the headlong desire to slash controls without considering the strategic cost. We are now suffering the consequences of this wrong-headed policy, and it is time for Congress to reverse it.

And fifth, the Subcommittee should examine the way the current export controls on computers were arrived at. The controls are based on a seriously flawed study commissioned without competitive bidding by the Commerce and Defense Departments. It is obvious from the study that its recommendations for decontrol are not supported by its findings. The Subcommittee should ask to see all the drafts of this study, and all the comments on the drafts by other federal agencies. The drafts and the comments make it plain that the current controls are devoid of any scientific basis. They are simply the product of a political desire to put trade above national security. Recently, the Commerce and Defense Departments have commissioned a follow-on study by the same consultants, with the objective of decontrolling computer exports even further. The Subcommittee should exercise its oversight powers immediately before another flawed study is produced.

Appendix to Testimony

In response to the requirement of Rule XI, Clause 2(g) of the Rules of the House of Representatives, the following information is provided describing contracts between the federal government and the Wisconsin Project on Nuclear Arms Control during the past three years

U.S. Arms Control and Disarmament Agency $24,500

U.S. Arms Control and Disarmament Agency $24,500

U.S. Arms Control and Disarmament Agency $10,000

U.S. Department of Energy $15,000

U.S. Defense Technology Security Administration $ 2,240

U.S. Department of Commerce BXA/DES $ 2,800

U.S. Customs Service $ 3,675

U.S. Department of the Treasury FAC $ 245

U.S. Department of Commerce BXA/DES $ 245

U.S. Arms Control and Disarmament Agency $ 245

Federal Bureau of Investigation $ 245

U.S. Department of Energy $ 485

Joint Warfare Analysis Center $ 245

U.S. Defense Intelligence Agency $ 397

National Air Intelligence Center Library $ 485

Sandia National Laboratory $ 485

Los Alamos National Laboratory $ 850

Since 1985, Professor Gary Milhollin has directed the Wisconsin Project on

Nuclear Arms Control, which carries out research and public education designed to

inhibit the spread of nuclear weapons. It operates in Washington, D.C. under the

auspices of the University of Wisconsin, where Professor Milhollin has been a

member of the law faculty since 1976.

Professor Milhollin holds degrees in engineering and law, and was an

Administrative Judge, part time, at the Nuclear Regulatory Commission for fifteen

years. He has practiced international corporate law in New York and Paris, and has

taught courses on nuclear arms proliferation at Princeton University and the

University of Wisconsin. He has also been a consultant on nuclear non-proliferation to the

Department of Defense. He testifies frequently before Congress

and his research has appeared in publications such as the New Yorker, the New York

Times Magazine and the Washington Post. Professor Milhollin is interviewed,

cited and quoted widely in the international press.

The Wisconsin Project's strategy has been to discover and publicize

clandestine nuclear transactions and the weaknesses in international agreements or

national laws that allow them to happen. Through its research reports, articles and

work with the press, the Project has influenced the export policies of major supplier

countries. The Project is funded by ten private foundations including the Carnegie

Corporation, the Ford Foundation and the W. Alton Jones Foundation.

APPENDIX A TO TESTIMONY OF GARY MILHOLLIN

Attached as Appendix A is an advertisement by Silicon Graphics, Inc. for one of its powerful new workstations. The advertisement, which shows pictures of Napoleon Bonaparte and Genghis Khan, says: "Napoleon worshiped it. Genghis Khan killed for it. Now you can buy it."

In the text, the advertisement says: "With a tool like this there could be yet another person looking to dominate the world."

The question this advertisement raises is whether Silicon Graphics has a responsible attitude toward the possible use of its products.

APPENDIX B TO TESTIMONY OF GARY MILHOLLIN

There is now pending before the U.S. Department of Commerce an application by Silicon Graphics to increase the power of a high-speed computer that Silicon Graphics exported to Hong Kong in December 1996 (Case No. D236990).

U.S. exporters are now required to obtain an export license before shipping a computer to Hong Kong if the computer operates above 10,000 MTOPS (million theoretical operations per second). The limit for mainland China is 2,000 MTOPS.

The computer that Silicon Graphics shipped to Hong Kong in December was configured to operate at 8,840 MTOPS, which allowed it to be sold without a license because its speed was just under the licensing threshold. Silicon Graphics almost immediately applied for permission to upgrade the computer to 11,770 MTOPS, which must have been its intended speed from the beginning. The computer is located at the Chinese University of Hong Kong's High Performance Computing Center, and will be available to its various departments of Physics, Chemistry and Engineering.

On July 1, 1997, Hong Kong will become part of mainland China, and therefore the computer will be at the service of the Chinese government in Beijing.

The Commerce Department should be required to state whether it intends to approve the export license application, and if so, what will be done to prevent the computer from falling into the hands of the mainland Chinese government.

APPENDIX C TO TESTIMONY OF GARY MILHOLLIN

In response to testimony given at this hearing by the Department of Energy, I would like to make the following comments on the extent to which the Silicon Graphics computers exported to Chelyabinsk-70 have increased the computing power available to Russia's nuclear weapon designers:

First, Mr. Viktor Mikhailov, Russia's Minister of Atomic Energy, declared to the press in Moscow in January 1997 that the Silicon Graphics machines increased the computing power available to Russia's nuclear weapon designers by a factor of ten.

Second, the increase in computing power cited by the Department of Energy's testimony is an estimate. It was a low estimate chosen from among several other higher estimates that were made by experts within DOE. Some of the higher estimates judged that Russia's computing speed had increased by a factor of thirty to a factor of one hundred.

Third, the method that DOE used to arrive at its estimate is not credible. It assumes that virtually all the available personal computers at Chelyabinsk-70 using Pentium processors would be taken off designers' desks and be hooked up in parallel with the most advanced software available. This is not a realistic scenario, and it does not compare the speed of the Silicon Graphics computers to the computers that Chelyabinsk was actually using before the export occurred. If such a comparison were made, the increase would be at least a factor of ten.

Fourth, DOE has a conflict of interest that should disqualify it from making an impartial estimate. Silicon Graphics is furnishing an important part of the computer power that DOE is using to maintain the U.S. nuclear weapon stockpile without underground testing. DOE officials in charge of Defense Programs have expressed the fear that if Silicon Graphics should suffer the penalty of being barred from the stockpile maintenance program, the program could be disrupted.