GPS and the Limits of Control

Right now, a 31-satellite constellation orbits Earth in medium Earth orbit, 20,000 kilometers above us. Each spacecraft is broadcasting a signal with the same strength as that of a residential light bulb. By the time it reaches Earth, that signal has attenuated to a whisper. Miraculously, these whispers are heard by 6 billion users every day and can be used to pinpoint your location to within three to five meters. Many of us cannot imagine a world without the Global Positioning System (GPS). Likewise, many may not remember that until 2000, civilian GPS was accurate only to distances about the length of a football field, a deliberate restriction imposed by the U.S. government.

The Air Force first developed GPS in the early 1970s to provide global position, navigation, and timing to American military forces. A GPS receiver calculates location by measuring the travel time of radio signals that arrive from GPS satellites that circle Earth broadcasting time and position information. For decades, concerned that adversaries would use GPS to guide weapons systems, the Pentagon degraded the accuracy of civilian GPS through a policy called Selective Availability.

But at midnight on May 1, 2000, the Clinton administration halted Selective Availability, giving the public free, unaltered access to GPS. Why?

To answer this question, we examined the Clinton administration’s decision through three lenses: commercial, technical, and political motivations. We found that on each axis, the answer lies in the speed of the American innovation ecosystem. While GPS was built by the military, commercial, civil, and military innovators adapted and improved it faster than policy could keep pace. Even with Selective Availability in place, commercial GPS device manufacturing outpaced military procurement; engineers invented technical workarounds to mitigate signal degradation; and, with the original constellation barely finished, the Pentagon started demanding a modernization program. The pace of innovation led the Clinton government to this critical policy outcome. The ideas of a new technology rapidly outpacing policy to govern it might seem familiar to some readers in the age of artificial intelligence (AI). In July, more than 1,200 employees of frontier AI companies signed a statement called “Pacing the Frontier.” In it, they ask the U.S. government for “technical and governance tools needed to deliberately pace the frontier of automated AI development.”

The story of Selective Availability shows how the speed of American innovation never relents, despite policymakers’ attempts to control where emerging technology goes. It is an illustrative case study in dual-use technology governance—and a preview of the developments policymakers may expect when they seek to control frontier AI.

Commercial Benefits

The first and most visible place in which American innovation outpaced Selective Availability was the commercial market. In the 1980s, despite the degraded accuracy of civilian GPS service, American companies began building devices based on it at a considerable pace, before GPS was even fully deployed, for navigation, surveying, mapping, and tracking vehicles, ships, and other assets. In fact, the proliferation of commercial GPS receivers would become essential for the military, too, in a striking example of authentically dual-use technology.

Early evidence of the vitality of the commercial positioning market came during the first Gulf War in 1991. Navigation proved challenging in desert combat, so GPS was essential for U.S. and allied forces to determine their exact position in featureless desert terrain. However, there was a severe shortage of military GPS receivers, with only 550 PSN-8 Manpacks—one of the first portable GPS receivers for combat troops—available. Very quickly, troops began using commercial GPS receivers that suffered from degraded position accuracy due to Selective Availability. Soldiers reportedly would ask their parents to mail them commercial GPS receivers. The commercial GPS market at the time included models such as the 1988 Magellan NAV 1000, used for recreational boating and hiking and sold for $3,000. After receiving their care packages containing these receivers—which, remember, had limited capabilities due to Selective Availability—soldiers would duct tape them to their Humvees for navigation guidance. As demonstrated by solders’ reliance on commercially produced GPS receivers, the commercial user equipment industry had clearly outpaced military receiver production. By 1994, the number of civilian GPS receivers in military use was in the tens of thousands. This left the Defense Department in an undesirable position. The U.S. military was dependent on commercial equipment but was simultaneously degrading its accuracy through Selective Availability. To leave Selective Availability in place was to leave U.S. forces at a disadvantage of their own making.

The Gulf War episode illustrates the velocity of American commercial enterprise (including GPS manufacturers such as Garmin, Trimble, and Magellan) in contrast to military acquisition programs. When the Gulf War broke out, GPS was not fully operational, and commercial manufacturers were already selling the same products to a consumer base of both outdoor enthusiasts and the families of U.S. soldiers stationed in Kuwait. The gap between commercial and government ability to scale new technology was the first sign that Selective Availability would become untenable.

At the same time, the rapid emergence of the commercial GPS industry called attention to the burden Selective Availability imposed. A 1994 market analysis prepared by Booz Allen Hamilton for the National Research Council estimated that the market for GPS receivers and systems would reach $42 billion over the period of 1994 to 2004 with Selective Availability in place, versus $64 billion if it were removed.

The Clinton administration’s policies leading up to the removal of Selective Availability articulate a clear goal of expanding civil and commercial GPS use. When Clinton first indicated his intent to discontinue Selective Availability in a 1996 Presidential Decision Directive, he committed the United States to encouraging “private sector investment in and use of U.S. GPS technologies and services.” The same directive established a permanent Interagency GPS Executive Board (now the National Executive Committee for Space-Based Positioning, Navigation, and Timing), jointly chaired by the departments of Defense and Transportation, to manage the system. Vice President Gore put the administration’s position plainly when he announced the civil GPS modernization initiative in 1999, describing GPS as a tool “to help guide everything from planes, trains, ships, and cars to tractors, snowplows, earthmovers, and mining equipment.” When Clinton announced his decision to remove Selective Availability in 2000, he valued the current market for GPS applications at $8 billion and projected that it would double within three years.

Indeed, the market grew exponentially. In the 26 years since the removal of Selective Availability, the commercial GPS economy has exceeded Clinton’s prediction. Every ride-hailing trip, every food delivery, every turn-by-turn direction, every tagged photograph runs on the accurate civilian signal. When you call 911 from a mobile phone, dispatchers can locate you because your phone broadcasts a GPS-derived position. Financial transactions depend on precise GPS time stamps for regulatory compliance and fraud detection. Precision agriculture uses GPS to guide autonomous tractors along rows. In other words, the removal of selective availability has benefited both industry and government: Law enforcement can complete lifesaving missions quickly and efficiently, federal regulators have a clearer picture of our financial infrastructure, and farmers can more effectively harvest crops that the U.S. economy—and the average U.S. dinner table, for that matter—depend on. The Department of Commerce estimates that GPS has generated $1.4 trillion in U.S. economic benefits.

Technical Advances

The commercial market was not the only arena in which American innovation was outpacing Selective Availability. Both civilian and military engineers were dismantling its technical basis simultaneously. More specifically, Selective Availability became technically irrelevant in the mid-1990s due to the rapid emergence of two new technologies: civilian augmentation systems to thwart it and navigation warfare systems to replace it.

GPS provides two kinds of service, Standard Positioning Service and Precise Positioning Service. Before Clinton removed Selective Availability, Standard Positioning Service had a horizontal accuracy of 101.2 meters, while Precise Positioning Service had an accuracy of 16.4 meters. Access to precise positioning, tightly controlled by the Defense Department, was not available to civilians.

GPS works by comparing a locally generated copy of a satellite’s coded radio signal with the received signal to measure the signal’s travel time. Multiplying that travel time by the speed of light gives the range to the satellite. GPS satellites broadcast precise timing and ephemeris data—that is, each satellite broadcasts the orbital parameters that describe its location—so that the receiver can determine the satellites’ positions. The original GPS ranging codes were the coarse acquisition (C/A) code and the precision (P) code.

Standard Positioning Service used only C/A, which GPS satellites transmit over the L1 frequency (1575.42 MHz). Precise Positioning Service also incorporates P, which is transmitted by satellites on both the L1 and L2 (1227.60 MHz) frequencies. C/A repeats every millisecond, which allows any GPS receiver to acquire it rapidly. Meanwhile, P repeats every seven days, making it harder to acquire; P-capable receivers must know its phase before they can lock on. “Phase” here means position within the code sequence. A receiver does not simply read the GPS code as a message. It generates its own copy and slides that copy in time until the two line up, and the shift required to align them is the travel time. Sliding blindly works when the sequence is short. C/A has only about a thousand possible code positions, so the receiver can blindly test them all in a fraction of a second. P has roughly 6 trillion possible alignments, so the receiver must be told where in its weeklong code segment the broadcasting satellite is, before it begins generating its own copy.

The Department of Defense controlled access to the Precise Positioning Service by using Selective Availability and encrypting the P-code (as P(Y)). Selective Availability worked by injecting pseudorandom error into the satellite signals. It added a clock dither, which introduced errors to timing accuracy. It also introduced errors to the ephemeris data, misrepresenting the transmitting satellite’s position. Authorized Precise Positioning Service users had the ability to subtract the errors and recover full accuracy. Everyone else, including adversaries, civilians, and commercial industry, had to use the degraded signal.

Selective Availability turned out to be only partially effective in the technical sense. Innovators across academia, the private sector, and civil government all implemented augmentation systems to work around Selective Availability, exploiting the fact that receivers near one another experience nearly identical errors. A receiver at a precisely surveyed location can compare where GPS says it is to where it actually is and then broadcast the difference as a correction. Nearby receivers subtract that correction and recover most of the accuracy that Selective Availability was designed to deny. Real-Time Kinematic techniques pushed further, using the carrier wave itself rather than the code to reach centimeter precision for surveyors. The Federal Aviation Administration (FAA) Wide Area Augmentation System (WAAS) placed reference stations at surveyed sites around the country, which routed their measurements to master stations that computed corrections and broadcast those corrections from geostationary satellites. The FAA awarded the $475 million WAAS development contract in 1995. Thus the Department of Transportation began spending federal money at scale to reverse errors that the Department of Defense was deliberately inserting. These systems made Selective Availability increasingly unproductive and irrelevant.

At the same time, Defense Department technologists developed a more direct answer to the problem Selective Availability was meant to solve. Remember, the government’s original fear was that an adversary would use an accurate GPS signal to guide precision weapons against American forces. By the early 1990s, Air Force engineers developed GPS jamming systems that could deny the GPS signal in a defined theater without degrading the worldwide signal for allies and commercial users. Thus, navigation warfare was born. As Arthur Money, then assistant secretary of defense for command, control, communications and intelligence, stated plainly at the press conference announcing Selective Availability’s end: “The Department of Defense, I believe, has demonstrated the capability to negate GPS signals in a threat area, consistent with military needs and the President’s policy; thus, we now can set selective availability to zero. Given the widespread use of GPS for peaceful purposes, we believe this approach is more effective than worldwide degradation.”

There was, however, a complication that jamming alone could not solve. Because C/A and P(Y) share the L1 center frequency, a jammer powerful enough to deny GPS to an adversary 200 miles away might also flood the spectrum with enough power to jeopardize P(Y) for nearby American forces. The P code already performed significantly (10 decibels) better than the C/A code in the presence of hostile jamming. Still, navigation warfare was becoming more sophisticated and the military needed a more secure signal.

The solution? M-Code. Rather than concentrating energy at the center frequency shared by C/A and P(Y), M-Code uses a binary offset carrier modulation that places its power in two lobes flanking the center frequency. The split-spectrum architecture allowed the U.S. to fortify the military signal and spectrally separate military and civil codes. A jammer could saturate the center frequency, suppressing civilian signals, while M-Code sat safely on either side.

In short, technical innovation drove the removal of Selective Availability. Selective Availability had been outflanked by new civilian augmentation systems, replaced by novel precision jamming as the tool of adversary denial, and superseded by a new military signal architecture. Selective Availability could not—and did not need to—persist in the context of these new technologies.

Political Motivations

Technical developments made it feasible to remove Selective Availability. Yet those same developments made the Pentagon intent on modernizing GPS to replace the effects of Selective Availability—and they needed substantial funding to do so.

As Selective Availability became technically obsolete, GPS modernization grew increasingly important. If sophisticated users could already undo the U.S. military’s degradation, the U.S. military needed to change its security model: grant everyone access to GPS, selectively jam adversaries as needed, and develop a less-jammable code (M-Code) for itself.

But developing jamming techniques and its own secure code, which requires launching new satellites and retrofitting military platforms with that code, is expensive. Implementing the control software for M-Code, for example, costs billions of dollars across upgraded satellites, control systems software, receivers on military platforms such as aircraft, ships, and missiles, and more. Just the GPS Next Generation Operational Control System (OCX) program, built to manage M-Code, eventually cost $6.3 billion (before the Space Force canceled the program in April 2026). GPS had just become fully operational, and a project whose costs would eventually rival GPS was already necessary.

Justifying a new military program in the mid-1990s was a difficult sell. It was during this period, in 1993, that then Deputy Secretary of Defense William Perry hosted “the last supper.” Over dinner at the Pentagon, he told the CEOs of major U.S. defense contractors that the end of the Cold War necessitated shrinking defense budgets and industry consolidation. And the “peace dividend,” popularized by President George H.W. Bush during the dissolution of the Soviet Union, championed shifting federal funding from military budgets toward programs benefiting the U.S. economy. These events led to flat or declining military budgets throughout the 1990s.

A 1997 memorandum of agreement on “Civilian Use of GPS” suggested that the Defense Department was tying a signature peace dividend initiative to its own GPS modernization priority. The Department of Defense and Department of Transportation agreed that, as the Transportation Department funds a second coded civil GPS signal, the Defense Department would “investigate the development of an enhanced, exclusive military capability.” The document goes further, stating that the civil C/A code “will not be activated on L2 until a new military capability is operational.” In other words, in return for gradually expanding access to GPS to bolster economic development and scientific research, the Pentagon received a mandate to begin its GPS modernization efforts.

The year prior, President Clinton announced his intention to remove Selective Availability within the decade in a 1996 Presidential Decision Directive. The directive encouraged “acceptance and integration of GPS into peaceful civil, commercial and scientific applications” and “private sector investment.” Yet the same directive mandated the development of “measures to prevent the hostile use of GPS and its augmentations to ensure that the United States retains a military advantage.” By opening the undegraded signal to the world, Clinton also opened the door for the Pentagon to build something more strategic. The directive positioned military GPS upgrades and navigation warfare advances as necessary parts of unlocking civilian GPS, a politically expedient way to provide the Pentagon with its mandate for GPS modernization. The civilian GPS economy became the political impetus for modernization, driving the next generation of military investment, which led to the development of M-Code. The Pentagon responded to and leveraged civilian momentum to advance its own innovation cycle.

Lessons for Artificial Intelligence

The story of Selective Availability is one about control over new technology and its limits. GPS was originally viewed as too important—and too dangerous—to be freely available. Yet innovation by civilian and commercial entities made total government control of the technology untenable. By 2000, companies and scientists had proved the value of the civilian signal; augmentation systems had undone much of the degradation; navigation warfare offered a way to deny GPS in a theater rather than across the planet; and the Pentagon, needing a justification for modernization at the height of the peace dividend, found one in commercial growth. In the end, not even the government wanted Selective Availability left in place. Commercial, technical, and political forces were all expressions of the same underlying force: the speed of American innovation—fast, distributed, and ever persistent.

Despite its undoing, Selective Availability was not a failure. It would have been a failure not to address legitimate fears about GPS misuse. The restriction stayed in place for the years it took to create other GPS security mechanisms. Its impermanence reflects that, alongside the restriction, the government remained committed to building better systems to replace it (M-Code, navigation warfare) and to adapting to new technologies that undermined it (differential GPS). The problem it was built to solve never went away, but Selective Availability provided an interim pacing solution, showing that restrictions are useful when a dual-use technology is new.

Today’s frontier AI models can execute cyber exploits, contribute to bioweapons development, conduct fraud and influence operations at scale, and be used for other harms. Selective Availability provides a case study for AI governance. It decayed for technical reasons embedded in its design that were discoverable in advance, because the clock dither and ephemeris errors degraded the civilian GPS signal in predictable ways, rather than withholding access outright. In the context of AI, model safeguards similarly degrade model performance in certain areas without removing the underlying capability. For example, Anthropic’s Fable model routes certain cyber- or biology-related questions (about pathogen production routes or requests to write working ransomware code, for example) to a less capable model rather than answering them directly—degrading performance while leaving the underlying capability intact.

Meanwhile, the encryption of the P(Y) code, which effectively denied nonmilitary users access to GPS, was not so easily compromised. Selective Availability also imposed a global policy when its intent was to prevent GPS use in very specific theaters. Once theater-scale GPS denial became possible, the broader policy was unnecessary. Employees of frontier AI companies’ request in “Pacing the Frontier” is for global limitations in AI development. The GPS precedent suggests that while that may be useful in the near term, in the long term a more sustainable framework might be targeted denial: know-your-customer requirements, refusal of service to identified actors, and monitoring at the point of access.

Admittedly, Selective Availability for GPS is not a perfect analogy to today’s restrictions on the use of frontier AI. Unlike GPS, which was developed and controlled by the government at inception, there is no single program office or identifiable chokepoint for AI. Perhaps only a few labs have models competing to be the very best, but more than a dozen across the world have models that are not far behind. The timescales are also moving far faster: GPS remained largely the same for more than a decade, while AI models are more capable practically every month. As a result, when the Commerce Department attempted to enact one of the first forms of AI regulation by placing export controls on Anthropic’s Fable model, it did so suddenly, without warning, and other companies were unaffected. The controls were removed after only 19 days. The pace of AI development, its progress controlled by the private sector, and the number of actors involved make AI governance much harder to address.

Still, like GPS, AI governance must face the fact that the speed of commercial innovation can leave old governance regimes behind. Selective Availability ended when technical, political, and commercial factors together provided a succession path. Accuracy restriction was an interim measure; its retirement stemmed from government funding, technical innovation, and commercial market forces. Today’s AI safeguards lack that arrangement. Blunt restrictions are in place at the point of model access, and research into potentially more durable solutions (for example, robust evaluations, verified compute, attribution at the point of use, or hardware attestation) proceeds on the budget and timescale of a side project. Congress and federal agencies should prioritize these efforts as frontier labs prioritize model development, while recognizing that researching new governance regimes may constantly be necessary if the speed of innovation renders old ones obsolete.