Monday, 12 January 2004
Wiretapping & VoIP
Last week, Uncle Fed (specifically, the Department of Justice, the FBI, and the Drug Enforcement Administration (DEA)) asked the FCC to force providers of voice-over-Internet protocol (VoIP) services to provide easy "wiretapping" capability to federal and local authorities. See Declan's report on C|Net: "Feds seek wiretap access via VoIP." A few comments are in order before the press mangles this situation and manages to obscure the facts. (Not to impugn Declan; I thought his article was good.)
Lawyers are in the language business, so we should examine the word wiretap to shed some light on exactly what Uncle Fed is asking for. Webster's Dictionary defines wiretap as an intransitive verb meaning "to tap a telephone or telegraph wire in order to get information." This definition is too circular to be useful at first, but this circularity becomes important later. Dictionary.com's nominal definition is a better starting point: "A concealed listening or recording device connected to a communications circuit." This was an accurate physical description when the term arose, during electric telegraphy's youth.
In those days, telegraphic circuits were hard-wired — that is, each pair of telegraph stations was connected by a single wire with one operator at each end. (Busy pairs of stations were connected by multiple wires, each one having operators at both ends.) Each transmission wire was plugged into a magnet-driven apparatus at each end that translated incoming electric signals into audible sounds and generated outgoing electric signals when the operator pressed a button. For an excellent beginner's text on early telegraphic technology and the economic and cultural developments it spawned, see Tom Standage, The Victorian Internet (1998).
In this environment, police had two options for surreptitious surveillance: (1) force the operator to disclose a message's contents after he received it, or (2) intercept the signal between the stations. Option 1 was inefficient because it was slow (the police had to wait for someone else to translate the message from Morse code and deliver it to them), and operators could not always be trusted to keep surveillance secret. Therefore, laws were passed that made option two mandatory. Telegraph companies were required to cooperate with the installation of a device (the "tap") onto their transmission wires that allowed the police to siphon off a tiny amount of the electric signal between two stations and send that signal to a police-operated station.
Later, switching technology made telegraphy more flexible. A switching device made temporary connections between transmission wires coming into the telegraph station. This allowed one operator (or more, at busy stations) connected to the switch to monitor several incoming wires simultaneously. Wiretap devices evolved in lock-step with switches and were quickly moved inside the switches so that fewer taps could monitor more transmissions without being physically reinstalled over and over. Whether this new configuration continued to qualify as "tapping" a "wire" is debatable. Early switching devices made temporary physical connections between telegraph wires by means of a third wire. Early switch tapping devices siphoned the electric signal off this switching wire, so there is a plausible argument that the term was still an accurate physical descriptor. Today we would understand the tapping devices as monitoring the operation of the switch device, not an individual wire within the switch. While wiretapping remained a reasonably good logical description of the tapping device's function, its accuracy as a physical descriptor was highly questionable.
The point to take from this is that wiretap first became an ambiguous term more than a century ago. Now reconsider Webster's circular definition, "to tap a telephone or telegraph wire in order to get information." Webster probably intended to denote the tapping of a circuit, not a wire, but we can forgive lexicographers for not being electrical engineers. However, Webster's definition unambiguously means eavesdropping on a single transmission or group of transmissions between two specified end points. In my experience, this is how law enforcers, laymen, and journalists all use the term. To convey the idea of collecting more than this information, they use such words as surveillance, eavesdropping, or data sniffing.
If the introduction of circuit switching made wiretap an ambiguous term, then the introduction of packet switching renders it positively useless. Packet switching is the transmission technology underlying the Internet Protocol, which is used for all Internet (and most local area network (LAN)) transmissions. Packet switching involves breaking data down into tiny pieces ("packets") and sending each packet across the network individually. This system eliminates the need for circuit switching, which dedicates a circuit to each transmission for the duration of that transmission. Few transmissions use the circuit continuously, so circuit switching inevitably involves inefficient "down time" for active circuits. Consider, for example, how frequently people pause while talking on the telephone. No information is transmitted during these pauses, but their circuit is monopolized nonetheless. Other callers cannot use this circuit until the first call ends — which forces the phone company to install a sufficient number of circuits to carry the maximum foreseeable number of transmissions simultaneously. This extra infrastructure is expensive to install and maintain.
Packet switching allows a small number of circuits to accommodate many transmissions because each one uses the circuit only while information is being actively sent. During each pause, the circuit is used for other transmissions. Additionally, different packets from the same transmission often take different routes across the network. Intermediate nodes will send packets along different routes to bypass busy sections of the network to avoid delays, among other reasons. Since packets must reach the destination individually, it must contain complete addressing information so that intermediate nodes can route it appropriately.
The same features that make packet switching more efficient than circuit switching also make it cheaper. (Sarcastic aside: This is as close to a "law" as the "science" of economics can offer us.) They also make it much more difficult to monitor communications. By definition, packets of information do not all travel through a packet-switched network by the same route. Therefore, there is no central box inside which to install a tapping device, as there is in circuit-switched networks.
The good news for law enforcers is that there does exist a place where all packets of a transmission must pass through before they are dispersed. That place is wherever the sender connects to the Internet backbone. "Backbone" is the name for high-speed networks that carry most Internet data until that data gets very close to its destination, at which time it is moved to a smaller (and usually private) network. All packets must travel from the sender's computer to the backbone through some identifiable means of transmission, be it in a cable or via wireless transmission in a form such as Wi-Fi.
The bad news for law enforcers is that each computer (or network) that connects to the Internet is connected via its own "pipe." They must install "tapping" devices on the connection used by each individual computer whose users' communications they intend to monitor. This requires that they get much closer to the target of the surveillance than they did with circuit-switched networks. In the old days, they could install tapping devices inside the switch at the telephone company's office. Conceivably they might do something similar at the target's Internet service provider (ISP). The FBI's (since-renamed) Carnivore project was an example of this. Unfortunately, this arrangement monitored traffic from all the ISP's customers, not just the intended surveillance target. In order to separate the target's transmissions from everyone else's, Carnivore has to read all packets that pass through. The only real solution to this problem is to install a device very close to the target — for example, in the cable that physically connects him to his ISP or at the antenna via which he transmits information to his ISP. This poses two main problems. First, the target may notice an unfamiliar device outside his house or office and become aware of the surveillance. Second, it is expensive because the police need to build many more devices and pay officers for the time it takes to install them at disparate locations.
By now, the linguistic difficulty of referring to any surveillance of data transmitted via the Internet as "wiretapping" should be obvious. At this point, I would like to shift direction slightly and briefly address a few related problems.
First, it is far from clear that the FCC has the authority to regulate VoIP as if it were a telecommunication service. It was widely reported last October that a federal judge in Minnesota ruled that VoIP companies provide "information" services, not "telecommunication" services, which means that states cannot regulate them under the Telecommunications Act of 1996. On the other hand, the 9th Circuit ruled earlier that month that the FCC erred in classifying cable broadband as an "information" service rather than a "telecommunication" service.
Second, according to Declan, Uncle Fed wants the FCC to require VoIP providers "to rewire their networks to guarantee police the ability to eavesdrop on subscribers' conversations." This is technically possible only for a few such services. In my understanding, Vonage sells black boxes that take input from a telephone and transmit data through the user's broadband ISP connection to Vonage's network, where Vonage routes it to another Vonage device or to a circuit-switched telephone network. Therefore, Vonage may be able to install devices that "tap" a specified user's conversations. Other services, however, operate in a fundamentally different way. Skype, for example, does not have any communications network at all. Its client software transmits voice data using the same decentralized P2P architecture found in Kazaa, the popular file-sharing client. (Skype was, after all, designed by the makers of Kazaa.) Therefore, Skype has no capability to install tapping devices, even if it wanted to cooperate with a hypothetical FCC order.
Third, as discussed above, to surveil transmissions on a packet-switched network, the police must read all data packets that pass through. If they ignore any individual packet, they may miss a piece of the message they intend to intercept. This makes it an unavoidable certainty that any "packet sniffer" will collect data that is not legally subject to surveillance — it would exceed the scope of all but the most expansive warrants. (Never mind that any warrant so expansive is probably unconstitutional because it would fail to state with particularity the information intended to be collected). Depending on the environment where the sniffer is installed, it may also collect data transmitted by third parties, who are not the intended targets of surveillance and who have a reasonable expectation of privacy in their communications. This is a Fourth Amendment problem of enormous magnitude — one that is well beyond the scope of this weblog.
Fourth, Uncle Fed's own statistics for 2002 show that about 80% of all wiretaps — both federal and state — were for criminal investigations in the course of enforcing drug laws. Only the remaining 20% were used for all other types of investigations. One is left to wonder whether the alarmist language in Uncle Fed's letter to the FCC was disingenuous: "criminals, terrorists, and spies (could) use VoIP services to avoid lawfully authorized surveillance." Uncle Fed tries to make it sound as if wiretaps are already an effective tool against such people when his own statistics show that wiretaps are rarely used against them. It would be another matter entirely if Uncle Fed intended to use VoIP monitoring technology to enforce drug laws. Even then, none of the dope dealers I knew of in college even knew what "broadband" meant — so it was unlikely that any of them had the equipment necessary to use VoIP. Even if drug importers are more sophisticated, the police can still monitor their communications through conventional warrants and responsible police work.
In conclusion, the only thing I can really say is that Uncle Fed's request is problematic, at best — and I am just a guy with an interest in Internet law, not an expert in history, technology, or constitutional law. If Uncle Fed was trying to start a national debate on the merits of Internet surveillance, it is about time we had one. If he thought he could slip this in under the radar, shame on him.