Last week, I wrote a piece for Reason in which I took a close look at the USA PATRIOT Act’s “lone wolf” provision—set to expire at the end of the year, though almost certain to be renewed—and argued that it should be allowed to lapse. Originally, I’d planned to survey the whole array of authorities that are either sunsetting or candidates for reform, but ultimately decided it made more sense to give a thorough treatment to one than trying to squeeze an inevitably shallow gloss on four or five complex areas of law into the same space. But the Internets are infinite, so I’ve decided I’d turn the Reason piece into Part I of a continuing series on PATRIOT powers.  In this edition: Section 206, roving wiretap authority.

The idea behind a roving wiretap should be familiar if you’ve ever watched The Wire, where dealers used disposable “burner” cell phones to evade police eavesdropping. A roving wiretap is used when a target is thought to be employing such measures to frustrate investigators, and allows the eavesdropper to quickly begin listening on whatever new phone line or Internet account his quarry may be using, without having to go back to a judge for a new warrant every time. Such authority has long existed for criminal investigations—that’s “Title III” wiretaps if you want to sound clever at cocktail parties—and pretty much everyone, including the staunchest civil liberties advocates, seems to agree that it also ought to be available for terror investigations under the Foreign Intelligence Surveillance Act. So what’s the problem here?

To understand the reasons for potential concern, we need to take a little detour into the differences between electronic surveillance warrants under Title III and FISA. The Fourth Amendment imposes two big requirements on criminal warrants: “probable cause” and “particularity”. That is, you need evidence that the surveillance you’re proposing has some connection to criminal activity, and you have to “particularly [describe] the place to be searched and the persons or things to be seized.” For an ordinary non-roving wiretap, that means you show a judge the “nexus” between evidence of a crime and a particular “place” (a phone line, an e-mail address, or a physical location you want to bug). You will often have a named target, but you don’t need one: If you have good evidence gang members are meeting in some location or routinely using a specific payphone to plan their crimes, you can get a warrant to bug it without necessarily knowing the names of the individuals who are going to show up. On the other hand, though, you do always need that criminal nexus: No bugging Tony Soprano’s AA meeting unless you have some reason to think he’s discussing his mob activity there. Since places and communications facilities may be used for both criminal and innocent persons, the officer monitoring the facility is only supposed to record what’s pertinent to the investigation.

When the tap goes roving, things obviously have to work a bit differently. For roving taps, the warrant shows a nexus between the suspected crime and an identified target. Then, as surveillance gets underway, the eavesdroppers can go up on a line once they’ve got a reasonable belief that the target is “proximate” to a location or communications facility. It stretches that “particularity” requirement a bit, to be sure, but the courts have thus far apparently considered it within bounds. It may help that they’re not used with great frequency: Eleven were issued last year, all to state-level investigators, for narcotics and racketeering investigations.

Surveillance law, however, is not plug-and-play. Importing a power from the Title III context into FISA is a little like dropping an unfamiliar organism into a new environment—the consequences are unpredictable, and may well be dramatic. The biggest relevant difference is that with FISA warrants, there’s always a “target”, and the “probable cause” showing is not of criminal activity, but of a connection between that target and a “foreign power,” which includes terror groups like Al Qaeda. However, for a variety of reasons, both regular and roving FISA warrants are allowed to provide only a description of the target, rather than the target’s identity. Perhaps just as important, FISA has a broader definition of the “person” to be specified as a “target” than Title III. For the purposes of criminal wiretaps, a “person” means any “individual, partnership, association, joint stock company, trust, or corporation.” The FISA definition of “person” includes all of those, but may also be any “group, entity, …or foreign power.” Some, then, worry that roving authority could be used to secure “John Doe” warrants that don’t specify a particular location, phone line, or Internet account—yet don’t sufficiently identify a particular target either. Congress took some steps to attempt to address such concerns when they reauthorized Section 206 back in 2005, and other legislators have proposed further changes—which I’ll get to in a minute. But we actually need to understand a few more things about the peculiarities of FISA wiretaps to see why the risk of overbroad collection is especially high here.

In part because courts have suggested that the constraints of the Fourth Amendment bind more loosely in the foreign intelligence context, FISA surveillance is generally far more sweeping in its acquisition of information. In 2004, the FBI gathered some 87 years worth of foreign language audio recordings alone pursuant to FISA warrants. As David Kris (now assistant attorney general for the Justice Department’s National Security Division) explains in his definitive text on the subject, a FISA warrant typically “permits aquisition of nearly all information from a monitored facility or a searched location.” (This may be somewhat more limited for roving taps; I’ll return to the point shortly.) As a rare public opinion from the FISA Court put it in 2002: “Virtually all information seized, whether by electronic surveillance or physical search, is minimized hours, days, or weeks after collection.” The way this is supposed to be squared with the Fourth Amendment rights of innocent Americans who may be swept up in such broad interception is via those “minimization” procedures, employed after the fact to filter out irrelevant information.

That puts a fairly serious burden on these minimization procedures, however, and it’s not clear that they well bear it. First, consider the standard applied. The FISA Court explains that “communications of or concerning United States persons that could not be foreign intelligence information or are not evidence of a crime… may not be logged or summarized” (emphasis added). This makes a certain amount of sense: FISA intercepts will often be in unfamiliar languages, foreign agents will often speak in coded language, and the significance of a particular statement may not be clear initially. But such a deferential standard does mean they’re retaining an awful lot of data. And indeed, it’s important to recognize that “minimization” does not mean “deletion,” as the Court’s reference to “logs” and “summaries” hints. Typically intercepts that are “minimized” simply aren’t logged for easy retrieval in a database. In the 80s, this may have been nearly as good for practical purposes as deletion; with the advent of powerful audio search algorithms capable of scanning many hours of recording quickly for particular words or voices, it may not make much difference. And we know that much more material than is officially “retained” remains available to agents. In the 2003 case U.S. v. Sattar, pursuant to FISA surveillance, “approximately 5,175 pertinent voice calls .. were not minimized.”  But when it came time for the discovery phase of a criminal trial against the FISA targets, the FBI “retrieved and disclosed to the defendants over 85,000 audio files … obtained through FISA surveillance.”

Cognizant of these concerns, Congress tried to add some safeguards in 2005 when they reauthorized the PATRIOT Act. FISA warrants are still permitted to work on descriptions of a target, but the word “specific” was added, presumably to reinforce that the description must be precise enough to uniquely pick out a person or group. They also stipulated that eavesdroppers must inform the FISA Court within ten days of any new facility they eavesdrop on, and explain the “facts justifying a belief that the target is using, or is about to use, that new facility or place.”

Better, to be sure; but without access to the classified opinions of the FISA Court, it’s quite difficult to know just what this means in practice. In criminal investigations, we have a reasonable idea of what the “proximity” standard for roving taps entails. Maybe a target checks into a hotel with a phone in the room, or a dealer is observed to walk up to a pay phone, or to buy a “burner.” It is much harder to guess how the “is using or is about to use” standard will be construed in light of FISA’s vastly broader presumption of sweeping up-front acquisition. Again, we know that the courts have been satisfied to place enormous weight on after-the-fact minimization of communications, and it seems inevitable that they will do so to an even greater extent when they only learn of a new tap ten days (or 60 days with good reason) after eavesdropping has commenced.

We also don’t know how much is built into that requirement that warrants name a “specific” target, and there’s a special problem here when surveillance roves across not only facilities but types of facility. Suppose, for instance, that a FISA warrant is issued for me, but investigators have somehow been unable to learn my identity. Among the data they have obtained for their description, however, are a photograph, a voiceprint from a recording of my phone conversation with a previous target, and the fact that I work at the Cato Institute. Now, this is surely sufficient to pick me out specifically for the purposes of a warrant initially meant for telephone or oral surveillance.  The voiceprint can be used to pluck all and only my conversations from the calls on Cato’s lines. But a description sufficient to specify a unique target in that context may not be sufficient in the context of, say, Internet surveillance, as certain elements of the description become irrelevant, and the remaining threaten to cover a much larger pool of people. Alternatively, if someone has a very unusual regional dialect, that may be sufficiently specific to pinpoint their voice in one location or community using a looser matching algorithm (perhaps because there is no actual recording, or it is brief or of low quality), but insufficient if they travel to another location where many more people have similar accents.

Russ Feingold (D-WI) has proposed amending the roving wiretap language so as to require that a roving tap identify the target. In fact, it’s not clear that this quite does the trick either. First, just conceptually, I don’t know that a sufficiently precise description can be distinguished from an “identity.” There’s an old and convoluted debate in the philosophy of language about whether proper names refer directly to their objects or rather are “disguised definite descriptions,” such that “Julian Sanchez” means “the person who is habitually called that by his friends, works at Cato, annoys others by singing along to Smiths songs incessantly…” and so on.  Whatever the right answer to that philosophical puzzle, clearly for the practical purposes at issue here, a name is just one more kind of description. And for roving taps, there’s the same kind of scope issue: Within Washington, DC, the name “Julian Sanchez” probably either picks me out uniquely or at least narrows the target pool down to a handful of people. In Spain or Latin America—or, more relevant for our purposes, in parts of the country with very large Hispanic communities—it’s a little like being “John Smith.”

This may all sound a bit fanciful. Surely sophisticated intelligence officers are not going to confuse Cato Research Fellow Julian Sanchez with, say, Duke University Multicultural Affairs Director Julian Sanchez? And of course, that is quite unlikely—I’ve picked an absurdly simplistic example for purposes of illustration. But there is quite a lot of evidence in the public record to suggest that intelligence investigations have taken advantage of new technologies to employ “targeting procedures” that do not fit our ordinary conception of how search warrants work. I mentioned voiceprint analysis above; keyword searches of both audio and text present another possibility.

We also know that individuals can often be uniquely identified by their pattern of social or communicative connections. For instance, researchers have found that they can take a completely anonymized “graph” of the social connections on a site like Facebook—basically giving everyone a name instead of a number, but preserving the pattern of who is friends with whom—and then use that graph to relink the numbers to names using the data of a differentbut overlapping social network like Flickr or Twitter. We know the same can be (and is) done with calling records—since in a sense your phone bill is a picture of another kind of social network. Using such methods of pattern analysis, investigators might determine when a new “burner” phone is being used by the same person they’d previously been targeting at another number, even if most or all of his contacts have alsoswitched phone numbers. Since, recall, the “person” who is the “target” of FISA surveillance may be a “group” or other “entity,” and since I don’t think Al Qaeda issues membership cards, the “description” of the target might consist of a pattern of connections thought to reliably distinguish those who are part of the group from those who merely have some casual link to another member.

This brings us to the final concern about roving surveillance under FISA. Criminal wiretaps are always eventually disclosed to their targets after the fact, and typically undertaken with a criminal trial in mind—a trial where defense lawyers will pore over the actions of investigators in search of any impropriety. FISA wiretaps are covert; the targets typically will never learn that they occurred. FISA judges and legislators may be informed, at least in a summary way, about what surveillance was undertaken and what targeting methods were used, but especially if those methods are of the technologically sophisticated type I alluded to above, they are likely to have little choice but to defer to investigators on questions of their accuracy and specificity. Even assuming total honesty by the investigators, judges may not think to question whether a method of pattern analysis that is precise and accurate when applied (say) within a single city or metro area will be as precise at the national level, or whether, given changing social behavior, a method that was precise last year will also be precise next year. Does it matter if an Internet service initially used by a few thousands—including, perhaps, surveillance targets—comes to be embraced by millions? Precisely because the surveillance is so secretive, it is incredibly hard to know which concerns are urgent and which are not really a problem, let alone how to think about addressing the ones that merit some legislative response.

I nevertheless intend to give it a shot in a broader paper on modern surveillance I’m working on, but for the moment I’ll just say: “It’s tricky.”  What is absolutely essential to take away from this, though, is that these loose and lazy analogies to roving wiretaps in criminal investigations are utterly unhelpful in thinking about the specific problems of roving FISA surveillance. That investigators have long been using “these” powers under Title III is no answer at all to the questions that arise here. Legislators who invoke that fact as though it should soothe every civil libertarian brow are simply evading their responsibilities.