The Constitution You Did Not Know You Were Building

On revocable encryption, the Fourth and Fifth Amendments, and the difference between law that protects you and architecture that does

Keywords: Fourth Amendment, Fifth Amendment, First Amendment, code as speech, key disclosure, takings clause, due process, NFT, revocable encryption, digital ownership, Riley v. California, Carpenter v. United States, US v. Hubbell, Bernstein v. DOJ, Bluebook

There is a sentimental tradition in American constitutional commentary which treats the Bill of Rights as a finished document — engraved, glossed, and litigated to the point of near-completion — and which treats the digital revolution as a sequence of interesting puzzles for which the existing constitutional categories must be stretched, sometimes uncomfortably, to fit. The Fourth Amendment, on this view, was written for a world of physical houses and physical papers, and the question of what to do about smartphones and cloud storage is a matter of doctrinal extension. The Fifth Amendment was written for a world in which a man knew where his diary was, and the question of what to do about encryption keys is a matter of analogical reasoning.

This view is wrong, and it is wrong in a way that flatters lawyers at the expense of inventors. The Constitution is not a finished document. It is an architecture for the protection of certain liberties, and architectures are tested at their joints. The joints have been giving way for thirty years, and they have been giving way for a structural reason that nobody on the Supreme Court has named clearly, though Riley v. California, 573 U.S. 373 (2014), and Carpenter v. United States, 138 S. Ct. 2206 (2018), edge near it: namely, that the platform-mediated digital regime has placed a third party between the citizen and his property, and that third party is the vehicle through which every conceivable form of state intrusion has been routed.

The cure for this is not better case law. The cure is better architecture.

Consider the Fourth Amendment in its most basic posture. The Government, before searching your house, must obtain a warrant supported by probable cause and particular as to the place to be searched and the things to be seized. Berger v. New York, 388 U.S. 41 (1967). The Fourth Amendment, on its face, does not say that this requirement may be evaded by demanding that a third party — your custodian, your servant, your bookseller — produce your papers without telling you. The third-party doctrine, United States v. Miller, 425 U.S. 435 (1976), and Smith v. Maryland, 442 U.S. 735 (1979), said otherwise: when you voluntarily convey information to a third party, you assume the risk that the third party will reveal it to the Government, and the Fourth Amendment is unbothered.

The third-party doctrine made a kind of sense in the era of the bank teller and the telephone-company switchboard, because it described, in legal language, the limits of the citizen's reasonable expectation of privacy in records he had surrendered to others for processing. It made considerably less sense in the era of cloud computing, where the "third party" is an entity that holds the citizen's most intimate data because the architecture of consumer software has made it impossible to do otherwise. Carpenter recognised some of this when it held that the Government's acquisition of cell-site location information from a wireless carrier was a Fourth Amendment search, the seven-day record being functionally a tracking device. The opinion is narrow; its logic is not.

The deeper point that Carpenter gestures toward is that the third-party doctrine becomes incoherent when the third party is a forced intermediary. A man who chooses to tell his accountant about his finances has, in some intelligible sense, surrendered his expectation of privacy. A man who cannot store his data anywhere except on a platform has surrendered nothing voluntarily; he has been compelled by the architecture of his world to interpose a third party between himself and his own information. To say that this man has "voluntarily conveyed" his information to the platform is a fiction, and it is the kind of fiction that constitutional lawyers ought to be embarrassed by.

Now consider what the construction described in my previous essay does to this picture. The decryption capability for an encrypted asset is split across an owner-tier (held in a trusted execution environment that will not release it even to the owner) and an oracle-tier (held by pseudonymous, open-pool, on-chain-paid oracles who do not know each other and do not know the owner's real identity). Decryption requires both. There is no platform. There is no single custodian. There is no third party who holds the owner's data in the sense that the third-party doctrine contemplates.

This is not a clever workaround. It is a structural change in the legal landscape. The third-party doctrine cannot apply because there is no third party of the relevant kind. The pseudonymous oracle holds a piece of a key — a number with no semantic content, useless without the other pieces, encrypted in transit to the requester only. The TEE holds another set of pieces, but cannot release them, even to the owner. There is no party in the system who can comply with a Government subpoena to "produce the contents of the user's file" because no such party exists. The Government may subpoena every oracle in the pool; it will receive ciphertexts that are unreadable without the requester's own session key, and the requester is the citizen the Government is investigating.

This is the architectural answer to Smith v. Maryland. Not a doctrinal one — the case is what it is — but an architectural one. Smith described a world in which information voluntarily surrendered to third parties had no Fourth Amendment protection. The construction described here makes such surrender mathematically unnecessary. The citizen surrenders nothing of substance to any third party; the third parties hold pieces too small to be meaningful in isolation; the citizen's substantive privacy interest never leaves his own enclave. The third-party doctrine has nothing to operate on.

The Fifth Amendment problem is sharper, and the architecture answers it more cleanly than the courts have managed.

The classic puzzle: the Government has a suspect's encrypted laptop. The Government wants the password. The suspect refuses, citing the Fifth Amendment privilege against compelled self-incrimination. The courts have produced a doctrine that no one finds satisfying. The Eleventh Circuit, in In re Grand Jury Subpoena Duces Tecum, 670 F.3d 1335 (11th Cir. 2012), held that compelled decryption was testimonial because it required the suspect to use the contents of his mind, citing Doe v. United States, 487 U.S. 201 (1988). The Massachusetts Supreme Judicial Court, in Commonwealth v. Gelfgatt, 11 N.E.3d 605 (Mass. 2014), reached the opposite result under the foregone-conclusion doctrine of Fisher v. United States, 425 U.S. 391 (1976). The Pennsylvania Supreme Court split it down the middle in Commonwealth v. Davis, 220 A.3d 534 (Pa. 2019).

The doctrine is incoherent because the underlying activity — compelled decryption — is the kind of activity the Constitution did not anticipate, and the analogies the courts reach for fail. Is decryption like producing a key (not testimonial under United States v. Hubbell, 530 U.S. 27 (2000))? Is it like producing the combination to a wall safe (testimonial)? Doe itself acknowledged the line, with Justice Stevens dissenting that "a defendant can be compelled to produce material evidence that is incriminating . . . [b]ut it has, until today, also been clear that an accused cannot be compelled to use his mind to assist the prosecution in convicting him of a crime." 487 U.S. at 219.

The architectural answer cuts the puzzle differently. Compelled decryption is incoherent in the present construction because the citizen cannot decrypt without the cooperation of an open pool of oracles whose participation is gated by on-chain ownership state. If the Government has obtained the laptop, the Government can interrogate the TEE. The TEE will refuse to perform decryption operations because the on-chain ownership of the asset has not been transferred to the Government. The Government may demand that the citizen sign a transfer transaction; this is itself a testimonial act under Doe, requiring the citizen to use the contents of his mind to produce a particular cryptographic signature, and the privilege applies. The Government may attempt to compel the oracle pool; the oracles do not know the citizen's real identity, do not know which oracles will be selected by the per-request sortition, and cannot collectively be coerced because they are pseudonymous and dispersed.

The right way to think about this is that the Fifth Amendment privilege has been, in the platform era, a paper shield against a sword that did not aim at the citizen at all. The Government did not need to compel the citizen to decrypt; it could compel the platform to do so, and the platform's compliance was not the citizen's testimony. The architectural construction restores to the citizen the actual property the Fifth Amendment exists to protect: not merely the right to remain silent, but the structural condition that the citizen's silence is causally efficacious. In the platform regime, the citizen's silence yields no privacy because the platform speaks. In the architectural regime, the citizen's silence is sufficient because nothing else can speak for him.

This is, I submit, what the Fifth Amendment was always doing. It was protecting a structural relation — the citizen's exclusive control over his own incriminating knowledge — and the platform regime had quietly dissolved that structural relation while the courts argued over the doctrinal vocabulary.

Consider next the takings problem, which is the constitutional dimension that crypto-libertarians most often misstate.

The Fifth Amendment provides that "private property [shall not] be taken for public use, without just compensation." The doctrine of regulatory takings, Penn Central Transportation Co. v. New York City, 438 U.S. 104 (1978), and Lucas v. South Carolina Coastal Council, 505 U.S. 1003 (1992), holds that government regulation may effect a taking when it deprives the owner of all economically beneficial use of his property, or when it functionally appropriates a property interest, even without formal seizure.

There is a question whether digital assets — software licences, accounts, data stored on platforms — qualify as "private property" for takings purposes. The Supreme Court has not squarely held, but the dicta in Horne v. Department of Agriculture, 576 U.S. 350 (2015), and the recognition of intangible property rights in Ruckelshaus v. Monsanto Co., 467 U.S. 986 (1984), point in the direction of yes. The Federal Circuit and Court of Federal Claims have entertained the question seriously in cases involving software contracts; the answer remains unsettled.

What is settled is that the platform regime has made the question of takings nearly impossible to raise, because the platform itself, not the user, holds the relevant property right. A platform's decision to delete a user's account, suspend a user's licence, or modify a user's purchased content does not raise a Fifth Amendment claim against the Government because the platform is a private actor and the Government has not directly taken anything. The user has no property to defend; the user has only a contractual relationship with the platform, which the platform may modify at will under the terms of service the user clicked through and never read.

The architectural construction changes the property relation. When ownership of a digital asset is recorded on a public ledger, secured by a cryptographic key whose pieces are distributed across a TEE the citizen controls and a pool of pseudonymous oracles, the citizen owns something, in the sense that the law has historically required for property rights to attach. He has exclusive use, supported by the architecture rather than by the goodwill of a platform. He has the power to transfer, supported by the chain. He has the power to exclude, supported by the cryptography.

If the Government attempts to interfere with this property — by, for instance, compelling the oracle pool to refuse service, or by ordering the TEE manufacturer to revoke attestations, or by seizing the chain itself if such a thing were possible — the takings analysis suddenly becomes coherent. The citizen has a property right of constitutional dimension, and the Government has interfered with it; the question of just compensation is the natural next question, and the question of whether the interference was for "public use" within the meaning of Kelo v. City of New London, 545 U.S. 469 (2005), can be litigated meaningfully. None of this was possible under the platform regime, because the platform was the property-holder and the Government's pressure on the platform was the platform's regulatory burden, not the user's takings claim.

The architectural construction restores the citizen to the legal posture that the takings clause assumes. He owns something. He can lose something. The Constitution has something to protect.

The First Amendment dimension is more familiar to readers who lived through the cypherpunk wars of the 1990s, but it is worth restating because the doctrine is settled and the implications have not been fully absorbed.

In Bernstein v. United States Department of Justice, 176 F.3d 1132 (9th Cir. 1999), and again in Junger v. Daley, 209 F.3d 481 (6th Cir. 2000), the federal courts of appeals held that source code is speech for First Amendment purposes, and that government regulation of cryptographic source code was subject to First Amendment scrutiny. The Bernstein case in particular held that the Export Administration Regulations' restrictions on the export of cryptographic source code were a prior restraint on speech and unconstitutional under the standards of Freedman v. Maryland, 380 U.S. 51 (1965).

The implication for the present construction is direct. The cryptographic protocol — the threshold-secret-sharing scheme, the ECDH key derivation, the contestation Script — is speech. The trusted-execution-environment code that runs inside the citizen's enclave is speech. The oracle responses, encoded as on-chain transactions, are speech. Each component, considered in isolation, falls within the Bernstein line. The aggregate is, structurally, a system of communications between speaking parties, mediated by a public ledger that is itself a speech medium.

What this means in practice is that any government effort to suppress, license, or compel modification of the architecture must satisfy strict scrutiny. The Government may not require the TEE manufacturer to install a backdoor; this is compelled speech under Wooley v. Maynard, 430 U.S. 705 (1977), and a content-based restriction on the manufacturer's chosen technical expression. The Government may not order oracles to alter their attestation responses; this is compelled false speech, and there is no compelling interest that justifies it where less-restrictive alternatives exist. The Government may not prohibit the use of the protocol; this is a prior restraint on speech, the precise wrong that Near v. Minnesota, 283 U.S. 697 (1931), held the First Amendment exists to prevent.

The architectural construction does not merely protect the citizen's speech; it protects the speech of the system in a way that is much harder for the Government to reach. The citizen does not need to be a particularly committed speaker; he need only use the architecture, and the architecture itself does the speaking that the First Amendment protects.

There is a further constitutional dimension that bears mentioning, because it is the dimension in which the architecture and the law interact most subtly: the Due Process Clause of the Fourteenth Amendment, in its procedural guise.

The doctrine of procedural due process holds that the State may not deprive a person of life, liberty, or property without procedures appropriate to the deprivation. Mathews v. Eldridge, 424 U.S. 319 (1976), gives the standard test: weighing the private interest, the risk of erroneous deprivation, and the State's interest. In the platform regime, the relevant deprivation is typically a platform's decision to suspend a user's account, delete a user's content, or revoke a user's licence. The user has no due process claim against the platform, because the platform is not a state actor; the user has no due process claim against the State, because the State has not directly deprived him of anything.

This is a hole in the constitutional fabric that civil libertarians have been pointing at for fifteen years, and it is not getting smaller. The platforms have become functionally sovereign in their domains, dispensing or withholding access to digital goods that have become, for many citizens, essential to economic and civic participation. They operate without the procedural constraints that bind the State, and the State has no incentive to reach in and impose constraints that would limit its own influence over the platforms.

The architectural construction changes this calculation. When the citizen's ownership of a digital asset is recorded on a public ledger and secured by cryptography, no platform has the power to deprive him without going through the cryptographic procedure that the protocol defines. The "procedure" is the procedure: a transfer transaction, signed by the current owner, recording the change of ownership. There is no informal deprivation possible. There is no platform suspension that could function as a deprivation, because the platform does not hold the asset.

The citizen who wishes to challenge a deprivation has, for the first time in the digital era, a stable reference point: the public ledger record of his ownership. The deprivation, if it has occurred, has occurred through a recognisable cryptographic mechanism, and the dispute over its legitimacy can be litigated against that mechanism. The State, if it wishes to act against the citizen, must do so through procedures that the architecture exposes — and exposes publicly, because the chain is a public record.

What I am describing is not exactly a due process claim, in the technical sense, because there is still no state actor for the claim to run against. What I am describing is the structural condition under which due process becomes possible, because the deprivations are visible, recorded, and individually identifiable. The platform regime, in which deprivations were invisible and aggregated within private contracts, did not satisfy this structural condition. The architectural regime does.

A brief word, also, about the Contracts Clause and the Commerce Clause, because the architecture pulls on both in instructive ways.

The Contracts Clause, U.S. Const. art. I, § 10, cl. 1, prohibits any State from passing a law impairing the obligation of contracts. The Supreme Court has narrowed the clause considerably since Home Building & Loan Association v. Blaisdell, 290 U.S. 398 (1934), but the clause is not dead; Allied Structural Steel Co. v. Spannaus, 438 U.S. 234 (1978), preserved it as a meaningful limit on state legislation that retroactively alters private contract terms. The architectural construction is, at one level, a private contract written in cryptography rather than English: Alice and Bob, transferring an asset, are entering into a legally enforceable arrangement whose terms are mathematically encoded. State legislation that ordered the oracle pool to alter its responses, or required TEE manufacturers to subvert their attestations, would be retroactively impairing the obligation of every such cryptographic contract in force at the time of the legislation. The Contracts Clause has work to do here that it has not, in two centuries, been called upon to perform.

The Commerce Clause, U.S. Const. art. I, § 8, cl. 3, gives Congress power to regulate commerce among the several States. The federal authority to legislate over the architecture is real but limited. United States v. Lopez, 514 U.S. 549 (1995), and United States v. Morrison, 529 U.S. 598 (2000), reminded us that the commerce power is not a general police power; it reaches activities that substantially affect interstate commerce, but does not reach private cryptographic protocols whose connection to commerce is incidental. An open pool of pseudonymous oracles operating across jurisdictions presents a particularly difficult Commerce Clause question because the pool itself has no centre, no jurisdiction of incorporation, no point of regulatory grasp. This is, of course, the design point: architecture that resists regulation does not invite Congress to assert authority over it; it asserts its own authority over the question of whether such regulation is feasible at all.

I want to draw the threads together, because the picture from a sufficient distance is more interesting than any of its parts.

The Bill of Rights was written in a world of physical objects, physical persons, and physical jurisdiction. The protections it offered were structurally tied to the architecture of that world. The Fourth Amendment protected your house because your papers were in your house. The Fifth Amendment protected your silence because the contents of your mind were inaccessible without your speech. The Fifth Amendment's takings clause protected your land because the State could not occupy it without you noticing. The First Amendment protected your speech because speech was something a citizen did, not something a platform did on his behalf.

The digital revolution dissolved these architectures. Your papers ended up on a server in a state you have never visited. Your silence yielded no privacy because a platform spoke for you. Your property became contractual entitlements held by intermediaries. Your speech became a stream of data routed through corporate gateways. The Bill of Rights, applied to this dissolved architecture, becomes a set of doctrinal puzzles that the courts cannot resolve because the underlying conditions the doctrines presupposed no longer obtain.

The architectural construction described in these essays is one attempt — and not the only one possible — to reconstruct the conditions. The ownership relation is restored: the citizen has something, and that something is mathematically defined. The third-party intermediary is removed: the pieces are distributed across pseudonymous open pools that cannot be subpoenaed. The compelled-decryption problem is dissolved: the citizen cannot decrypt unilaterally and so cannot be coerced to do so. The takings claim is made coherent: the citizen has property to defend. The First Amendment claim is made strong: the architecture is itself speech, and pervasively so. The due process condition is established: deprivations are public and recorded.

I do not claim that any of this is a substitute for the Constitution. I claim that the Constitution, applied to a world that has the architectural conditions the Constitution presupposes, works; and that applied to a world that lacks them, it does not. For thirty years the lawyers have been trying to make the doctrine work in the absence of the conditions. They have failed, not because they are stupid, but because the task is impossible. The conditions must be rebuilt before the doctrine can run on them.

There is a phrase I am tempted to use, though it is grand, and the temptation is to be resisted in proportion to its grandness. We get the Constitution we are willing to architect. The framers did not create their structural protections by writing them into a document; they created them by living in a world that materially supported them, and by writing the document to memorialise what was already true. The platforms created their dominance not by litigating; they created it by building. The reconstruction of constitutional liberty in the digital age is, by the same logic, an engineering problem before it is a litigation problem.

The architecture I have described — revocable encryption, threshold secret sharing, pseudonymous on-chain oracles, contestable economic enforcement — is one fragment of one possible reconstruction. There are others. Some are better; some are worse. None of them will be built by the Supreme Court. All of them will be built, if they are built at all, by people who understand that the real protection of liberty in our age is going to require the same combination of cryptographic ingenuity and institutional patience that the framers brought to the original drafting.

The Constitution is not a document. It is an architecture. We have, accidentally and at length, dismantled most of it. We can rebuild it, with care, and the tools to do so are now available to anyone with the will to learn them.

That, in the end, is the whole of the matter.