On August 2, 1964, North Vietnamese torpedo boats had attacked the US destroyer Maddox in international waters as that ship was supporting raids on the North Vietnamese coast. SIGINT had warned of an attack, and the torpedo boats were repulsed. Score one for SIGINT.
Then, two nights later, in the midst of Johnson administration warnings to North Vietnam about further action, SIGINT misread North Vietnamese reporting on their continuing recovery operations from the first night as a second attack and issued a CRITIC (a kind of global warning). Subsequent US Navy evasive action and firing at some spurious radar hits were duly noted in North Vietnamese shore-based communications, which were in turn picked up by NSA and errantly catalogued as further evidence that a second attack was under way. President Johnson ordered air attacks against the torpedo boat bases, and Congress delivered the Gulf of Tonkin Resolution, authorizing the president to take “all necessary steps” to halt Communist aggression in Southeast Asia.
In the aftermath of the resolution, NSA stuck to its story that a second attack had occurred. It’s unclear if the agency’s subsequent investigation was careless, misguided, or just consciously ignored evidence. But it is clear that the August 4 reporting was wrong.
Tonkin and Iraq’s WMD were sobering lessons.
• • •
SURPRISINGLY, AS DIRECTOR OF NSA, I received no formal guidance to prepare for war with Iraq. But it didn’t take a rocket scientist to know that we were going to war, so we moved around what we later estimated to be about $400 million to get ready for the conflict. The deputy head of our SIGINT Directorate, a Gulf War veteran, directed the team we set up to prepare for war to read the agency’s history of the first Gulf War. He then conducted what the army calls a “rock drill” to synchronize what various tactical and national SIGINT units would do to support one another.
Linguists were a perennial challenge, even before the demands of this war. A year earlier, in 2000, with al-Qaeda strengthening and with information on their plans sparse, CIA had outfitted an early model of the Predator with electro-optical and infrared cameras to hunt al-Qaeda in Afghanistan. The agency was looking for bin Laden, his close associates, training camps, and any evidence of weapons of mass destruction. It was tough getting this under way; a whole infrastructure of people and equipment had to be built to handle the streaming video from the UAV (unmanned aerial vehicle).
Those early missions did not carry SIGINT packages, but that didn’t let NSA entirely off the hook. As weak as Afghan air defenses were, we still didn’t want the political embarrassment of losing a drone, so we could not simply ignore the Taliban’s limited radar coverage and its handful of MiG-21s. Any Afghan detection of the Predator was deemed sufficient to scrub the mission. So we couldn’t fly without monitoring the air defense network, and that required linguists. Actually, we were so limited in Pashto speakers, it required the linguist, who would be dedicated to this task whenever the Predator was in Afghan airspace.
The Afghans finally did attempt an intercept, but the Predator was so small and so slow that the unlucky MiG-21 pilot reported that he could not find it—a conversation duly reported by NSA. In truth, the actual intercept was a little oblique. Something along the lines of the “birds spotted a pigeon but lost it.” A good object lesson for anyone who thinks it’s OK to “talk around” classified data on the phone.
Great work, but we were going to have to get a whole lot better to support full-scale conflict or rather, as it turned out, several conflicts: Afghanistan, al-Qaeda, soon-to-be Iraq. We began to pull Arab linguists off mission in order to sharpen them in the Iraqi dialect. All accounts, except for CT, suffered.
There was no magic lever to pull here. Good language analysts are a rare and special breed, since they have to be masters of so many skills.
First and foremost, they need to know the language and the culture. They have to be target smart: Who is related to whom? What is the overall context? What code words do they use? Who lies to whom on the phone? Finally, they’ve got to know the SIGINT system. Intercepts don’t come to you on a platter. You have to be a hunter far more than a gatherer.
In ways not widely understood—taking a communication not in your language, between two individuals not of your culture and perhaps not quite able to distinguish the world as it is from the world as they would like it to be (common in messianic groups like al-Qaeda)—taking all of that and turning it into something actionable is high art and science.
In one instance, analysts were puzzled by the continued references to a specific household article from a known terrorist group. One linguist decided to stay at work for a weekend to listen intently to all the available cuts. Analysts usually scan cuts to get to the important ones; otherwise they would never get through the queue. But by giving up a weekend, this analyst listened to everything and gathered references to this household article and a variety of colored accessories. By Sunday night he had it. The household article symbolized passports. The accessories were visas. Their colors signified specific countries. Breakthrough.
The last time NSA intercepted bin Laden was in 1997, but that wasn’t the last time it heard him. In 2000, one AQ intimate of bin Laden’s was calling another and using a harsh tone, almost barking orders, in a way that seemed very inappropriate to the language analyst. The cut was puzzling. Why the odd tone? When the analyst finally turned up the volume to a very high level, there, identifiable in the background, was the voice of bin Laden himself telling the first communicant what to say.
Only a few days after 9/11, Russia passed us a warning about al-Qaeda possibly using “the big one” based on an intercept of a communication between two terrorists. We immediately thought nuclear and hunted throughout our system to find a copy of the cut. We succeeded, and luckily it was of very high quality. With George Tenet demanding answers fast, I put our very best linguist on it. Within thirty minutes he reported that the targets were certainly players in the terrorist network, but the question was whether or not they were talking about a nuclear attack on us or whether they were talking about our taking dramatic action against them. It was indeed speculation about our response to them. False alarm, this time.
So language skills really mattered. I upped the required standard for our military linguists. In an earlier world, where we were intercepting heavily formatted military messages, basic fluency would suffice. A ground controller directing a fighter to turn left or turn right or to ascend to a certain altitude was pretty easy to follow. Al-Qaeda didn’t talk that way. They were elliptical, metaphorical, indirect, nuanced . . . and clever. We needed better skills.
On the civilian side, in the four years after 9/11, we were able to hire about five hundred new linguists. We raised their pay too. A lot. Several times.
Some of the new hires were native-born in target languages and, since they were young, pretty tech savvy too. Others were older, desperate to serve, but they made it quite clear that they would never master the technical intricacies of the SIGINT system. They did make superb quality controllers on our translations.
There were special emotional strains in this line of work. Language analysts stay on a target for a long time and sometimes know their target better than they know their own families. They become part of the virtual life of the target. It can be a real emotional roller coaster. When a target has been designated for direct action and his death needs to be confirmed, it is often the same phone that is targeted, but now the communications comprise sobbing, grieving family members.
Some analysts have had real issues with confirming the identity of a communicant prior to direct action. Some just can’t do it. This is conscientious objection to killing or at least to being a direct part of the killing, not a political view on the cause or conduct of a war. One senior recounted to me one of his linguists spontaneously beginning to cry as she was in the car with him. He asked what was wrong. She said, “Something at work,” and when they got to the office, she
explained the emotional stress of targeted killings.
Special people. And rare. One day I was returning from a meeting at Langley and my security detail looked a little impatient, as they had to wait for a stream of pedestrian traffic flowing outward from the NSA headquarters building. As they inched the SUV forward, I only half in jest cautioned, “Be careful. They could be linguists.”
Before the Iraq war started, I wanted to talk to the entire workforce. The congressional vote on authorizing the invasion promised to be close, and I had every reason to believe that the workforce split along similar lines. I usually gave an important briefing once in the agency’s Friedman Auditorium and then had it repeated and beamed electronically to workstations at Fort Meade and beyond.
Not this time. Too important. I wanted a personal touch. I gave the briefing to multiple audiences (clearly not to all 35,000 military and civilian workers, but you get the point) at Fort Meade and up near the Baltimore-Washington airport, where we had a large annex. Because we had folks working with Central Command in Tampa and in the combat theater, we knew CENTCOM commander Tommy Franks’s concept of operations—“shock and awe”—and the broad outline of his war plan.
I walked the workforce through our responsibilities, how our emphasis would shift as the assault force went through its phases:
Phase 1. Deployment.
Phase 2. Shaping the battle space.
Phase 3. Major force-on-force action.
Phase 4. Post-conflict stability operations.
Knowing the political divisions in this country and the upcoming Senate debate, I acknowledged that there were surely some in the audience who opposed the coming conflict. So I reminded them of our duties as professionals, that if the republic authorized war, we would fight it savagely, limited only by the laws of armed conflict. I then added, “We could all agree that it would be a bad thing indeed if countries around the world got the idea that it was OK to be an enemy of the United States of America.”
I was asked in several meetings how long all this was going to take. Don Rumsfeld had gotten the same question while visiting the US air base at Aviano, Italy. He said, “It is unknowable how long that conflict will last. It could last six days, six weeks. I doubt six months.”
I had a little different answer. I said that the part that the American people would call war, phase 3, would be measured in weeks, not months. I then added that phase 4, post-conflict stability operations, would not be measured in months either. I got the WMD wrong, but I certainly nailed that one.
Even before Franks ordered his forces to cross the line of departure, NSA was already deeply in the fight, so some at the agency were surprised at my reaction to the traditional signs that NSA used at moments like this: “We support the warfighter.”
“Don’t say that,” I said. “We’re not support. We’re part of the fight. We’re responsible for final outcomes, not just our inputs.”
I had concluded that we had to go forward with combat formations. When you are there, you are in the fight. When you are not, you are in the in-box, and no one reads their in-box when they are in a fight.
The major muscle movement before us was transitioning from being a SIGINT production factory to an intelligence consultancy. We couldn’t just mail it in and merely sustain the old transactional relationship of delivering SIGINT products. We needed a living, breathing operational intimacy with people we used to call customers. We had to ramp up the forward deployment of our knowledge, skills, and abilities. One senior summarized it as sending our carbon units rather than our silicon units into the fray.
I thought that we were doing well enough until, after the war was under way, our chief of research, the one from Walt Disney Imagineering, took an extended trip through the theater. He reported that the new approach was working, to a point. “They love us at MNF-I [Multi-National Force–Iraq] and corps headquarters. We’re OK at division level, but below that they barely know who we are.” As director of the National Security Agency, doctrinally I was charged with supporting the national leadership and Department of Defense echelons down to about the corps level. But doctrine wasn’t working. Lower-echelon units weren’t feeling the love (or getting and acting on valuable SIGINT).
We reached further down-echelon and embedded five-person NSA teams at the division level and linked regiments (in the Marine Corps) and brigades (in the army) to the highly classified NSANet. Once we trained them and cleared them for very sensitive information, low-echelon GIs were getting access to precious national databases, all the while being advised by high-end NSA professionals.
In one telling vignette, an NSA analyst at Fort Meade reported that he had “received a secure call from a master gunnery sergeant in the front line. He needed to know what kind of switching equipment was used in An Nasiriyah’s central switch. Seems they were planning an operation against it and were interested in exploiting the switch, if they could get control of it. Fortunately, NSA had that information—it was an Alcatel E-10. I had a mental image of him sitting in the back of an armored vehicle, using a laptop to link into NSANet.”
When we really got going, front-line SIGINT soldiers and marines were tuning orbiting satellites to home in on targets to their immediate front, while folks under my direct control at Fort Meade and Fort Gordon in Georgia were tuning antennas on tactical vehicles in the forward line of march.
To keep all of this synchronized, we held routine videoconferences every other day for all the relevant SIGINT units inside and outside the theater. My conference room screen would have a dozen or more locations beamed in as we shared information, assigned tasks, and de-conflicted operations. The playbook was “coherent centralized planning [with] decentralized execution; a networked information sharing enterprise; [leveraging] an unprecedented level of collaboration.”
To be clear, we weren’t trying to play Tom Sawyer in order to get a lot of other people to whitewash our fence. In fact, it was more the opposite. We welcomed more help but had to be careful whom we trusted with a brush.
Even though making SIGINT is hard work, a lot of people think they can do it. When customers think they’re not getting enough of NSA’s attention—or when they are, but the reporting is not what they want to hear—the agency sometimes gets a request for the “raw data.” The request usually betrays a real lack of understanding of how this works.
“How raw do you want it? Before we process it, when it’s still unintelligible beeps and squeaks?”
“No. No. After processing.”
“Sure.”
“But this is all in Urdu or Pashto or something.”
“Yes, I know.”
“But I need it in English.”
“We don’t translate and store everything in English. Just the important stuff.”
“Yeah. Give me that stuff.”
“But that’s what you have been getting. We call it reporting.”
On the rare occasion when we turned a raw SIGINT pipeline on, we usually received a request within hours to turn it off. Too much stuff.
We were being theologically correct (and consistent with our charter) to keep what we called the SIGINT production chain under our control. But we were truly dealing with a ton of material, and we could never be totally sure that we completely (or currently) understood our teammates’ requirements. And frankly, people everywhere were now participating in processes that had always served them, but from which they had previously been personally barred, from teller machines to booking airline tickets to publishing books.
We needed to square this circle, and we hit upon the concept of the SIGINT production stream, our routine reporting at the mouth of our waterway and raw unprocessed traffic at the headwaters. We would allow people and organizations previously known as customers to swim upstream as long as they could protect the data they encountered and as long as they could add value to it. Those two caveats weeded out the casual
request and the ill-considered demand.
The further upstream someone wanted to go, the heavier the requirement. “Protecting the data” meant secure spaces, secure communications, and secure people (read polygraphed). That was quite an investment. “Add value” meant that their efforts wouldn’t simply result in more noise in the system. What they were doing had to be worthwhile. That required training (like protecting US privacy) and skills (like language).
We weren’t going to provide a SIGINT peep show, but I also made clear to the NSA folks that the value added by these customer-producers didn’t have to match the value we thought we could add. If they thought it worth their effort, we would not stand in the way. We had plenty of work to do, anyway.
• • •
OVERALL, WE DIDN’T DO BADLY. A few days into the Iraq conflict we could report that the Iraqis were having trouble communicating and had evacuated several facilities; power outages were affecting some units and SAM sites had been damaged.
But we weren’t getting the flood of signals we had anticipated. Chinese firms had put down extensive fiber-optic cable after the first Gulf War, and it was clear that the Iraqis were riding those lines now. We stepped up pressure on CENTCOM planners to put cable heads on the target list and to hit them hard even though some were in or near populated areas. The idea was to herd signals into the air, where we could intercept them.
It worked; kinetic operations were enabling intelligence collection. A week into the conflict, signals spiked to such a degree that linguists had to be called in off break to handle the increased volume. That greatly facilitated locating Iraqi units for destruction. Two weeks later we reported that the SIGINT targets we had been following in southern Iraq and Baghdad had “largely ceased to exist.”
The early bombing campaign had been bedeviled by GPS jammers that the Iraqis were using to spoof the guidance on American smart bombs. The SIGINT system managed to locate these elusive devices, and they were added to the target list as well. We also intercepted the communications of technicians servicing the system and cleared our reports to support State Department démarches to the supplier country.
Playing to the Edge: American Intelligence in the Age of Terror Page 7