Between Edward Snowden’s ongoing leaks and a series of frankly unprecedented disclosures by the government itself, the public now knows quite a bit about the NSA’s controversial telephony metadata program, which makes use of the Patriot Act’s §215 to collect, in bulk, nearly all Americans’ domestic  call detail records from telephone carriers.  We know far less, however about the government’s bulk collection of Internet metadata under FISA’s pen register/trap-&-trace authority, which supposedly ceased in 2011—though some such collection almost certainly continues in a more limited form. That collection merits closer attention, because the legal argument that bulk metadata acquisition doesn’t violate the Fourth Amendment—rehearsed in a recent post by Orin Kerr—simply doesn’t work for acquisition of Internet metadata at the backbone, for technical reasons that it’s not at all clear the Foreign Intelligence Surveillance Court has considered.

The FISC’s recently declassified memorandum opinion authorizing bulk telephony metadata collection contains a dismayingly cursory Fourth Amendment analysis resting on the now-familiar reasoning of Smith v. Maryland: Users voluntarily convey phone dialing information to a “third party” (i.e. the phone company), knowing that information will be retained in the company’s records for routine business purposes.  They thereby “assume the risk” that these records will be shared with the government—notwithstanding any contrary promises of confidentiality—and so waive their Fourth Amendment expectation of privacy in that information.  This is the so-called “third party doctrine.” The ruling in Smith has been widely and justly condemned—and as Jennifer Granick has ably argued here, is of dubious relevance to NSA’s bulk collection program anyway. But let’s pretend for the moment, strictly arguendo, that this reasoning is not crazy on its face.

We often see Smith glossed as standing for the proposition that, to borrow Orin’s phrasing, “metadata… is not protected under the Fourth Amendment.” What’s important to recognize here, though, is that this isn’t strictly accurate. The government may not, for instance, break into my desk drawer to obtain my old phone bills on the theory that such information is categorically unprotected, however and wherever obtained. Rather, Smith turns on an argument about one’s expectation of privacy in information knowingly conveyed to a provider and routinely stored in its business records when the government obtains such information from that provider. (The dialing information in Smith was actually obtained in realtime via a  “pen register” device installed at the phone company’s central offices, rather than via business records, but what matters is that the government was still obtaining from the phone company the very information routinely recorded and retained by the phone company for its business purposes.) The Court in Smith explicitly distinguished such records from the contents of a telephone communication, which still enjoy Fourth Amendment protection, even though those call contents are conveyed through the provider, and the provider will typically have the technical capability to intercept and record such content.  One may still “assume the risk” that the other party to the call will disclose its contents to law enforcement, but callers are not imagined to assume any such risk vis a vis the phone company itself, even though the carrier is capable of disclosing those contents to the government as well. Hold that thought for a moment.

Returning to the newly released FISC opinion, we find reference on page 8 to another case in which that court had reason to analyze the bulk collection of metadata, and similarly found no Fourth Amendment obstacle to the collection, seemingly relying on the same reasoning from Smith v. Maryland. A brief passage from that earlier opinion takes up the question of whether the enormous volume of collection makes a difference, concluding that “[s]o long as no individual has a reasonable expectation of privacy in meta data [sic], the large number of persons whose communications will be subjected to the … surveillance is irrelevant to the issue of whether a Fourth Amendment search or seizure will occur.”

A bit later, on page 20, the FISC again references prior rulings on bulk collections “different from the one at issue here,” this time explicitly citing the FISA pen register provision, 50 USC §1842(c)(2). It seems reasonably likely that these are references to either the same opinion or a series of related opinions validating bulk collection of Internet metadata.  It is, of course, impossible to tell for sure from such brief excerpts, but the quoted passage suggests that the FISC may be misreading Smith as standing for the proposition that “meta data” as such lacks Fourth Amendment protection.

That’s potentially significant because an enormous amount of public reporting—stretching back to whistleblower Mark Klein’s 2006 disclosure of a “secret room” where NSA filtered essentially alI Internet traffic routed by AT&T—suggests that NSA is pulling at least some of that metadata directly off the Internet backbone via equipment installed at the facilities of Internet Service Providers. Indeed, for communications merely transiting through the United States—such as an e-mail sent from one foreign provider to another—this might be the only place to acquire the relevant metadata. It’s also the most obvious reason to use a pen register order, since records held by providers like Google and Yahoo could also presumably be obtained via bulk §215 orders of the same sort that are served on telephone carriers.

The trouble, as I explained in a recent article for Ars Technica, is that the Internet functions quite differently from the traditional circuit-switched telephone network. On the phone network, a binary distinction between “content” and “metadata” works well enough: The “content” is what you say to the person on the other end of the call, and the “metadata” is the information you send to the phone company so they can complete the call. But the Internet is more complicated. On the Open Systems Interconnection model familiar to most techies, an Internet communication can be conceptualized as consisting of many distinct “layers,” and a single layer may simultaneously be “content” relative to the layer below it and “metadata” relative to the layer above it.

To make this concrete (and oversimplifying slightly): Suppose you send an e-mail to me, the user jsanchez at the server, over a broadband connection provided by Comcast. Relative to the contents of your message, the e-mail header containing the sender, recipient, and other information is “metadata.” But Comcast wouldn’t normally have any reason to “look” at that header information, let alone store it in a  “business record,” in order to route the data properly—any more than Verizon would normally eavesdrop on the part of a phone call where you’re asking a receptionist at a corporate switchboard to connect you with a particular party. The ISP is just moving packets from one IP address to another—from your laptop to your e-mail server (or maybe not, if you happen to run your own e-mail server), and then from your e-mail server to Cato’s. All the ISP normally needs to “look at” to do its job is that IP address layer.  And the IP addresses it processes might be uniquely associated with a particular domain (like, but could just as easily be shared among several domains. Only once those packets arrive at their destination does Cato’s e-mail server need to look at the e-mail header information to determine which inbox to drop the message in. Indeed, the only way Comcast would have that metadata is by illicitly intercepting the contents of my communications with Cato’s e-mail server. I could tell a similar story, though in some cases differing significantly in the details, for Web traffic, IM chats, peer-to-peer file transfers, and other forms of Internet communication.

The crucial point here is that the detailed “metadata” for a particular Internet communication, past the IP layer, typically wouldn’t be processed or stored by the ISP in the way that phone numbers and other call data is stored by the phone company. From the ISP’s perspective, all of that stuff is content. Depending on the particular communication, those further layers of  metadata might be stored as business records by some other “third party” service provider, like Google—or they might not.

Either way, the acquisition of “metadata” other than IP addresses from an ISP or off the backbone is pretty clearly dissimilar from the collection of call data at issue in Smith in every important respect. It is not information conveyed to the Internet provider for the purpose of routing the communication; it is routing information conveyed through the provider just like any other content. Nor is it information the Internet provider would otherwise normally retain for routine business purposes. Again, relative to the ISP, it’s all just content.

Some of that content, I noted above, might ultimately be processed and retained as a business record by some other company—as, indeed, the contents of an e-mail or phone call might. And following the logic of Smith, those business records might then be obtained from those companies without a Fourth Amendment warrant. But none of that should matter to the Fourth Amendment analysis of acquisition at the ISP or backbone. As Justice Scalia wrote for the Court in Kyllo v. United States:

The fact that equivalent information could sometimes be obtained by other means does not make lawful the use of means that violate the Fourth Amendment. The police might, for example, learn how many people are in a particular house by setting up year-round surveillance; but that does not make breaking and entering to find out the same information lawful.

Similarly, the government might obtain from a hotel’s switchboard logs some record of which particular guest I asked to be connected with when I called—but this does not make it any less a Fourth Amendment search to wiretap my conversation with the concierge. The government might be be able to subpoena a list of my e-mail correspondents from Google or Cato—but that doesn’t make it any less a Fourth Amendment search to obtain the same data by seizing my laptop or tapping my broadband connection.

If this seems like nitpicking—who cares exactly where they get the data, once we’ve acknowledged they can get it legally from someone or other?—consider some concrete cases. Suppose I’m engaged in correspondence with a reporter for the Associated Press, and both of us are exclusively using our work accounts. The metadata records of that correspondence will normally exist only on Cato’s e-mail server and AP’s. Perhaps neither of us can vicariously assert a Fourth Amendment right against the compelled production of those server records. Nevertheless, as the FISC opinion notes, the record holder itself is free to do so. And in many circumstances, it’s quite plausible that both Cato and the AP would be prepared to contest an order for the production of those records on First and Fourth Amendment grounds. Comcast or AT&T, on the other hand, would not have much incentive to do so, and so it could easily make quite a big difference if, instead of obtaining records from Cato or the AP, the government is able to collect that information by selectively eavesdropping on the contents of the communications between their e-mail servers.  More generally, Internet companies like Google and Yahoo have demonstrated substantially greater willingness than telecom providers to fight against government requests they regard as overbroad. Interception of metadata on the backbone, then, ensures that neither the end users nor the e-mail providers have any opportunity to assert their rights.

As a practical matter, this point seems to be largely moot in ordinary criminal investigations, because ordinary criminal investigators don’t have cutting edge semantic analyzers sitting on the Internet backbone. They’re investigating specific, known targets, and the easiest place to get a specific target’s e-mail metadata is, unsurprisingly, from their e-mail provider, where the familiar argument of Smith is, let’s say, no less plausible than usual.  It’s only for the kind of surveillance the NSA wants to do—identifying unknown targets through the bulk analysis of all metadata—that a pen register at the backbone would make sense. The only problem is, the legal analogy to a traditional telephone pen register completely breaks down once you look at the details.

We have not, of course, seen the FISC’s detailed legal analysis of bulk Internet pen registers—for all we know, the Court’s opinion includes a lengthy and sophisticated analysis of how the architectural differences between packet switched and circuit switched networks affect the applicability of Smith. But the language of the just-released §215 opinion does not fill me with confidence. Rather, the Court repeatedly writes as though a particular “type of information”—and indeed, perhaps anything characterized as “meta data”—can be categorically outside the protection of the Fourth Amendment, and thus obtained by any process described as a “pen register” without conducting a Fourth Amendment “search,” regardless of the technical details of how or from whom that information is obtained.  But since those very details did matter in Smith, it ought to be troubling if Smith‘s holding is being reflexively applied to a radically different fact pattern.

Since the FISC has already ordered the release of more legal opinions concerning the §215 program, this might be a good time to apply the same spirit of transparency to its rulings on pen/trap orders for Internet metadata.  If I have not given the FISC judges enough credit here, the public can be reassured by witnessing directly the breadth of their technical understanding and the thoroughness of their legal reasoning. If, on the other hand, their reasoning has not been so well attuned to the technical details, letting legal scholars and technologists alike check their work could be the first step toward correcting a significant constitutional error.