Two weeks ago I wrote about the problems for law enforcement created by Apple’s (and Google’s) new plans to offer widely available, effectively unbreakable encryption for iPhones, prompting strong negative reactions from FBI Director James Comey and others. I pointed out there that such a feature would create practical impunity in a wide variety of cases where individuals encrypt evidence of their illegal activities and render that evidence inaccessible even with legal authorization. The conversation has developed, with dueling legal proposals.

Last week, Comey gave a speech at the Brookings Institution in which he raised the possibility of requiring technical changes to the encryption scheme. The speech is worth reading in its entirety, but the most interesting section is his proposal that Congress update a 1994 law called the Communications Assistance for Law Enforcement Act (CALEA):

Current law governing the interception of communications requires telecommunication carriers and broadband providers to build interception capabilities into their networks for court-ordered surveillance. But that law, the Communications Assistance for Law Enforcement Act, or CALEA, was enacted 20 years ago—a lifetime in the Internet age. And it doesn’t cover new means of communication. Thousands of companies provide some form of communication service, and most are not required by statute to provide lawful intercept capabilities to law enforcement.
What this means is that an order from a judge to monitor a suspect’s communication may amount to nothing more than a piece of paper. Some companies fail to comply with the court order. Some can’t comply, because they have not developed interception capabilities. Other providers want to provide assistance, but they have to build interception capabilities, and that takes time and money.
The issue is whether companies not currently subject to the Communications Assistance for Law Enforcement Act should be required to build lawful intercept capabilities for law enforcement. We aren’t seeking to expand our authority to intercept communications. We are struggling to keep up with changing technology and to maintain our ability to actually collect the communications we are authorized to intercept.

Although Comey doesn’t get into details of the proposal, it is likely similar to a proposal that the FBI was offering early last year. That proposal, according to the Washington Post, wouldn’t set “rules that dictate how the wiretap capability must be built,” but instead would “let companies develop the solutions as long as those solutions yielded the needed data.”

What is new is the apparent suggestion (at least suggested by the context) that CALEA could be amended to require smartphone companies to let the government decrypt smartphones on demand. But this proposal runs squarely into the problems that I identified before.

First, smartphone engineers could create this capability by weakening the smartphone encryption scheme so it could be broken quickly. But any encryption scheme that could be broken quickly by law enforcement would necessarily be weak tea, and would also be breakable by a well-resourced attacker like a foreign intelligence service or a criminal organization. Such weak encryption would be especially problematic for companies that need to protect proprietary business information from prying eyes overseas.

Second, engineers could create a secret backdoor key reserved specially for the smartphone manufacturers themselves or the government, to be used only under the proper legal authorization. But this poses similar dangers. The backdoor key itself would become a valuable prize. In addition, a secret backdoor likewise introduces another point of weakness into the already-complex cryptographic system. As security expert and cryptographer Bruce Schneier has noted, “Building a secure cryptographic system is easy to do badly, and very difficult to do well.” Mandating an additional layer of complexity would complicate an already quite difficult technical feat, increasing the probability of errors and the likelihood of unintended consequences.

Seeking to do an end-run around these types of technical problems, George Washington University law professor Orin Kerr has proposed (for discussion purposes only) a purely legal solution. Kerr proposes that Congress define the new crime of “willful refusal to comply with a decryption order.”

Consider a thought experiment. What if Congress enacted a new statute that directly links punishment for the refusal to the punishment for the crime under investigation, so that the punishment for refusal to comply would be set as the punishment for the underlying offense?

In Kerr’s proposal, refusal to unlock a device in an investigation for a crime would carry the same penalties as the crime being investigated, which we might term the “reference crime.” This would, at least in theory, equalize the incentives for failure to comply, since he would be facing the same penalties for both unlocking or refusing to unlock.

For several reasons, I’m not convinced that this proposal would be a practical solution.

First, although the possible punishment for refusal to decrypt might be legally identical to punishment for the reference crime, it’s not clear that the punishment would usually be the same. If the government wants to decrypt a phone, it’s presumably because the phone is believed to contain important evidence. But the reason the government is still searching is the same reason that owner doesn’t want to decrypt: He knows what’s on the phone; the government doesn’t. If the owner knows that the government doesn’t know about evidence of other crimes on the phone, he won’t decrypt.

Second, there’s the trouble of determining which crime is the appropriate reference crime. If the government can just arbitrarily specify which crime was being investigated based on the possibility that the evidence might turn up something very serious, that would create a problem of due process and ex post facto lawmaking. That would also give the government an incentive to always pick the most serious possible crime, whether or not there is actual evidence of it, so as to create the biggest possible incentive for cooperation. If the reference crime is only based on whatever the government has probable cause for, then we’re back to the first incentive problem. As such, Kerr’s proposal would consistently either overestimate or underestimate the proper punishment.

Third, even if these problems were solved and the legal incentives were equalized, I’m still not convinced that the owner would ever want to decrypt. If given a choice between being convicted of, say, production of child pornography and refusal to decrypt, why would a subject ever choose the social stigma associated with the substantive crime?

Finally, all of this presupposes that the crime could still be investigated after giving the defendant an opportunity to obey the decryption order. If the tactic worked at all, it would be limited to situations when serving a decryption order wouldn’t immediately result in destruction of evidence, notification of co-conspirators, or hasty departure from the jurisdiction.

In sum, neither Comey’s legal-technical solution nor Kerr’s purely legal proposal is particularly attractive. Even assuming that a purely technical solution would be feasible—that is, government security researchers figuring out how to circumvent the encryption—it wouldn’t eliminate many of these problems. Apple and Google’s wholesale deployment of encryption to consumers definitely creates a serious ethical problem that legislation won’t do much to fix. But this problem is the inevitable price of helping consumers protect themselves in an increasingly dangerous digital world.