The Electronic Frontier Foundation has unearthed an interesting case of an improper use of surveillance in an investigation where the FBI had obtained “roving wiretap” authority. In a bizarre turn, the Bureau ended up eavesdropping on young children rather than their adult suspects for five days. The case is generating some attention because that same “roving wiretap” authority is one of the three surveillance powers set to expire in late May. The thing is, on the basis of what I can glean from the heavily redacted document EFF obtained via a Freedom of Information Act request, it’s not a case involving misuse of the roving authority. But it is a good concrete example of why the roving authority needs to be modified.

First, a bit of background: Roving wiretaps in criminal cases have been around since the 1980s, and were designed to let investigators follow a target across multiple telephone or Internet accounts when there’s reason to believe the suspect is frequently changing lines to frustrate surveillance. The Patriot Act extended the same power to intelligence investigations—fine in principle—but with a crucial difference. On the criminal side, there’s a tradeoff: For a conventional warrant, you have to specify every phone line and e‑mail address covered by a wiretap warrant, but as long as you’ve got probable cause to link each line to criminal activity, you don’t necessarily need the identity of target. For a roving criminal wiretap, you don’t have to be limited to phones and accounts identified in advance, but you do have to name a target, and then directly ascertain that this named target is “proximate to” every facility you’re about to start collection on. Having both is even better, but you must have one or the other: either a named, known suspect or a list of specific facilities. Not so in the Patriot Act version used for intelligence wiretaps: They only require a “specific” description of the target. Critics argue that these “John Doe” warrants fall short of the Fourth Amendment’s requirement that warrants “particularly [describe] the place to be searched, and the persons or things to be seized,” and worry that they dramatically magnify the risk of erroneous collection of innocent people’s communications, especially given that intelligence wiretaps are, in many ways, already much more broad—and sweep in a lot more communications—than their counterparts on the criminal side.

The report EFF obtained concerned an investigation in which, it appears, roving wiretap authority was granted by at least one of two orders authorizing surveillance of two subjects, implied to be a married couple. Because it’s so heavily redacted, it’s necessary to read between the lines a bit, but as best I can reconstruct it, here’s what appears to have happened. In the course of the surveillance, an analyst was surprised to find that the conversations on one monitored line sounded like they involved “young children” who were not speaking the same language as the targets of the investigation. The Bureau nevertheless continued monitoring it for five days. Eventually, someone realized that they’d already been informed that the targets had disconnected that phone line a while ago—at which point, presumably, it was reassigned to the kids the FBI ended up wiretapping. In fact, the Bureau had initially refrained from collecting on that line, because the techies had determined that the subjects of the investigation had already disconnected it by the time surveillance was ready to begin. When the request for a renewal of the surveillance order was made, however, the lawyer at DOJ who drafted the renewal request mistakenly included that old number in the application, even though the case agents hadn’t put it in their request, and seemingly noted explicitly that it wasn’t being used by the targets anymore. But it got into the renewal order, at which point whoever was charged with implementing it dutifully began tapping some child’s phone line.

On face, the improper collection here was not the result of a “John Doe” warrant. Again, it’s hard to be absolutely sure through all the redactions, but the wording of the report strongly suggests that both targets were known, named individuals from the outset. Neither does the mistake here seem to have been a direct result of the roving authority. The wording suggests that the phone number at issue was actually specified in the initial order, rather than being selected as an exercise of roving discretion by the case agents. Even if it had initially been identified that way, the Bureau realized the subjects weren’t using it anymore before they actually went up on the number… at least originally. So one might think this is just a case of an error with conventional intel wiretapping, with no obvious implications for the roving authority as such. But I think there are a couple of important lessons to draw from it.

First as the report notes, the lawyer who drafted the renewal application and the case agents who reviewed and signed off on it all dropped the ball in failing to notice that it included a number they’d already concluded wasn’t being used by their targets. The FISA Court itself, however, was not really in any position to detect the error. So while it is an unqualified improvement that roving authority, since 2006, has included a notice requirement to the court within ten days of a wiretap roving to a new number or account, we need to be realistic about how much that’s going to accomplish. If an investigation starts going up on some conspicuously excessive number of accounts, or adding them with explanations to the court that sound wildly implausible, you might see a judge step in. But for the most part, when a tap goes up on the wrong person, we’re going to have to count on the agents and analysts themselves to notice.

Second, this is a good illustration of an important difference between criminal wiretap orders and FISA warrants. With a criminal order, interception is “minimized” in realtime—meaning you wouldn’t have had five days of automatic recording of every conversation on the line. Someone would listen to the start of the call, and as soon as it became apparent they couldn’t positively identify a known target as party to the communication, or that it was a personal call unrelated to the crime under investigation, monitoring would (in theory) stop. There’s necessarily some flexibility in practice, but the default on the criminal side is not “record everything”; on the intel side, it is.

Third, consider how important a named and identified target is to quickly catching an error. In this case, one of the technicians responsible for executing the wiretap noticed that the people named in the warrant had already disconnected service to one of the targeted phone lines. Having an actual identity for your target is also a decent proxy for having many other types of information about them: In this case, the analyst monitoring the recordings had a set of expectations about what the subjects sounded like and what language they would be speaking. In other cases, it might become apparent that a phone is making calls from one location when the target specified in the warrant is known to be elsewhere. Could these things be known for a target whose name isn’t? Sure. But as a rule, a description specific enough to guide initial targeting isn’t necessarily data-rich enough to enable very robust error-correction. Since the presumption is that people wiretapped under FISA warrants do not get notified even after the fact, this internal correction is the only real check on erroneous targeting. When wiretaps are limited neither by facility nor a known target identity, it’s not just that there’s greater risk of tapping the wrong line—it’s that it becomes much more likely nobody will be able to detect with any certainty that a mistake has been made.