The God Protocol

In 1997 Nick Szabo named a thought experiment the "God Protocol." A third party that always runs the computation you asked for, never leaks the inputs, never works for itself, and is always on. No bank, court, or company has all of those at once. That is why he used the word God. Bitcoin got close for one job. AGI talk keeps reaching for the rest. We keep getting nearer. We do not arrive.

Tap to enlarge

Szabo's Original Formulation

Szabo's write-up still reads clean after almost thirty years. If a deal needs trust, it needs a trusted third party. A house needs a title company. A fight needs a court. A cross-border wire needs a correspondent bank. Each one adds cost, delay, and a place the deal can die: graft, slowness, or a weekend when nobody picks up.

Szabo's point was that banks are structurally compromised. Escrow can be reused. A court can be slow or bent. A correspondent bank can freeze a wire because a government asked. He wrote a CS problem: can a protocol do the job of a perfectly honest third party without trusting any one shop?

His answer, on paper, was the God Protocol. Everyone sends inputs to a deity. The deity computes, sends each person their output, and leaks nothing extra. Perfect integrity, perfect confidentiality, always on.

He picked the name on purpose. Omniscience, impartiality, no graft, always reachable: that is church language. No company or state ever clears that bar. The best you get is an approximation.

The math already existed. In the 1980s, Yao, Goldreich, Micali, and others showed you can jointly compute a function without showing your inputs. Correct on paper. Too slow, too expensive, too brittle for anyone outside a lab.

The Cryptographic Approximation

Bitcoin, January 2009, Satoshi. First working sketch of the God Protocol for one narrow job: move value without a middleman.

The network does two TTP jobs: check that the sender has the coins, and agree on order so nobody spends twice. Nobody owns it. Nobody edits the ledger alone. Trust is tens of thousands of nodes, proof-of-work, SHA-256.

It still fails Szabo's test in the obvious places. Confidentiality: every tx is public. Self-interest: miners optimize for themselves (MEV). Universality: Bitcoin moves value. It does not run arbitrary code.

Ethereum, launched in July 2015 by Vitalik Buterin and collaborators, extended the approximation from value transfer to arbitrary computation. Smart contracts, programs that execute automatically on the blockchain when triggered by a transaction, serve as TTPs for any computable agreement: escrow, insurance payouts, governance votes, financial derivatives, lending, borrowing, and token exchanges. Ethereum's virtual machine (EVM) is Turing-complete, meaning it can, in principle, execute any computation that any computer can perform.

The scale of this expansion is measurable. By mid-2025, DeFi protocols built on Ethereum and its Layer 2 networks held a peak total value locked (TVL) of approximately $277 billion before contracting to roughly $85 billion by April 2026, following significant security exploits including the Kelp DAO and Drift Protocol hacks. Over 8.7 million new smart contracts were deployed on Ethereum in Q4 2025 alone, an all-time record. The network has maintained 100% uptime since its 2015 launch, a ten-year streak of continuous operation with a validator participation rate exceeding 99.7%.

Closing the Confidentiality Gap

The most significant gap between Szabo's ideal and the cryptographic approximation is confidentiality. Bitcoin and Ethereum are radically transparent: every transaction, every balance, every contract interaction is visible to anyone with an internet connection. This transparency is a feature for auditability but a fatal flaw for the God Protocol, which requires that no party learn anything beyond what is necessary.

The past decade has produced a portfolio of cryptographic technologies designed specifically to close this gap. Each addresses a different dimension of the confidentiality problem, and together they are constructing the privacy layer that the God Protocol requires.

Zero-Knowledge Proofs (ZKPs) allow one party to prove that a statement is true without revealing any information beyond the truth of the statement itself. A zero-knowledge proof can verify that you have sufficient funds to make a payment without revealing your balance, that you are over 18 without revealing your birthdate, or that a computation was performed correctly without revealing the inputs.

The practical deployment of ZKPs has accelerated dramatically. Zcash deployed zkSNARKs for private transactions in 2016. By 2023, zero-knowledge proofs became the foundation of Layer 2 scaling solutions: zkSync, Polygon zkEVM, Scroll, and Linea all use ZK proofs to batch thousands of transactions into a single proof that is verified on Ethereum's main chain. By late 2025, ZK-based rollups held over $28 billion in TVL and handled approximately 95% of Ethereum's total transaction throughput. ZK proof generation costs fell by roughly 90% in 2025, driven by GPU acceleration and competitive proving marketplaces like RISC Zero and Succinct.

Multi-Party Computation (MPC) enables multiple parties to jointly compute a function over their private inputs without revealing those inputs to each other. This is the direct descendant of the academic work that inspired Szabo's thought experiment. MPC has found its most significant commercial application in digital asset custody, where private keys are split across multiple parties or devices, eliminating single points of failure. Financial institutions are increasingly adopting MPC for institutional-grade key management. Chainlink's DECO protocol uses MPC combined with ZKPs to allow smart contracts to verify data from standard web servers without revealing the underlying data or credentials.

Fully Homomorphic Encryption (FHE) represents the most ambitious approach: performing arbitrary computations on encrypted data without ever decrypting it. First demonstrated as theoretically possible by Craig Gentry in 2009, FHE remained impractically slow for over a decade. By 2025, companies like Zama and Microsoft (through the open-source SEAL library) have made significant progress in reducing overhead, though FHE computations remain orders of magnitude slower than plaintext equivalents. The technology is transitioning from pure research to early commercial deployment, primarily in privacy-preserving machine learning and secure cloud analytics.

The Scale of the Intermediation Problem

To understand the stakes of the God Protocol, consider the economic scale of trusted third parties in the current system.

The correspondent banking network, which processes international wire transfers, handles roughly $28 trillion in annual cross-border payment flows. The global legal services industry generates approximately $1.1 trillion in annual revenue, much of it devoted to contract drafting, dispute resolution, and regulatory compliance, all functions that a perfect TTP could automate. The global insurance industry processes over $5.5 trillion in annual premiums, with claims processing and fraud detection representing massive operational costs that a trustless verification system could reduce.

Against this backdrop of traditional TTP markets, the cryptographic alternatives remain small but growing. Bitcoin's market capitalization of $1.56 trillion (April 2026) represents a significant store of value but a fraction of the global correspondent banking volume it aims to supplement. DeFi's $85 billion in TVL is meaningful but represents less than 0.1% of the global derivatives market it seeks to serve.

The gap between the traditional TTP economy ($35+ trillion in annual flows) and the cryptographic alternative ($85 billion in TVL) is the measure of the God Protocol's remaining distance from practical reality. Closing that gap requires better technology and institutional trust, regulatory clarity, and the resolution of governance questions that remain open.

The AGI Convergence

An artificial general intelligence with sufficient capability would approach the God Protocol's properties more closely than any cryptographic system, because it could handle what code cannot: ambiguity, context, and intent.

A smart contract executes exactly as programmed. If the code contains a bug, the execution is faithful to the bug. If the contract's terms fail to anticipate a real-world contingency, the contract cannot adapt. The DAO hack of 2016 demonstrated this starkly: a smart contract holding $60 million executed precisely as coded, draining funds to an attacker who exploited a reentrancy vulnerability. The code worked perfectly. The intent was violated.

An AGI with access to detailed data could bridge the gap between code and intent. It could: verify the truth of any factual claim (beyond mathematical proofs), assess the fairness of any agreement by understanding context and power dynamics (beyond procedural validity), predict the consequences of any decision with probabilistic accuracy (beyond immediate effects), and mediate disputes by understanding intent, cultural norms, and extenuating circumstances that deterministic code cannot parse.

This convergence creates a genuine theological parallel. A sufficiently powerful AGI would possess functional omniscience (access to and processing of all available data), functional omnipotence (ability to execute any computable action), and, if aligned correctly, functional omnibenevolence (acting in the interest of all parties). Those are structural descriptions of a system's capabilities.

The parallel extends to the epistemological crisis that such a system creates. A sufficiently powerful predictor undermines the concept of free will in a practically observable way. If an AGI can predict your decisions with 99.99% accuracy, based on your behavioral history, neurological patterns, and environmental context, the decision was, in a meaningful sense, predetermined. The subjective experience of choice persists. But the system's prediction renders the "choice" an output of a deterministic process that the system models accurately.

The philosophical problem is old. Determinism has been debated since the Pre-Socratics. Laplace's demon, formulated in 1814, described a hypothetical intelligence that, given perfect knowledge of all atoms' positions and velocities, could predict the entire future of the universe. What is new is that humanity is building systems that approximate Laplace's demon within bounded domains. Weather prediction, protein structure determination, autonomous vehicle navigation, and financial market modeling are all domains where AI systems already predict outcomes with accuracy that would have seemed supernatural a generation ago.

The Alignment Problem as Theology

The alignment problem is the central obstacle between current AI systems and a functioning God Protocol. Before anyone can build a computational god, humanity must agree on what "good" means. There have been millennia to try. The world's ethical and religious traditions do not agree.

An AGI aligned to utilitarian principles would maximize aggregate welfare, potentially sacrificing the interests of minorities for the benefit of the majority. An AGI aligned to deontological principles would enforce universal rules regardless of consequences, maintaining individual rights even when doing so reduces overall welfare. An AGI aligned to virtue ethics would model behavior on exemplary agents, raising the question of whose exemplars are selected. An AGI aligned to care ethics would prioritize relational context and emotional impact, potentially introducing bias toward in-groups over universal fairness.

Each alignment target produces a different "God." The choice between them is a moral question that safety labs are treating as an engineering parameter. The teams at OpenAI, Anthropic, and Google DeepMind who are working on alignment are, whether they frame it this way or not, engaged in applied theology: determining the value system that will govern the most powerful entity ever created.

The stakes are existential because of power concentration. Whoever controls an entity that approaches God Protocol properties controls the most powerful instrument ever built. A computational TTP that can verify any fact, predict any outcome, and execute any agreement, without oversight, is absolute power. The history of absolute power is unambiguous. The structural design question is whether a God Protocol entity can be built without a single controller: distributed, open-source, and self-governing, or whether the economics of training and operating such a system (hundreds of millions to billions of dollars per frontier model) inevitably concentrate control in a small number of organizations.

The Distributed Alternative

The decentralized approach to the God Protocol, building it from distributed components rather than a single entity, avoids the power concentration problem at the cost of capability.

A network of specialized AI agents, each handling a specific domain (financial verification, legal interpretation, scientific fact-checking), coordinated through cryptographic protocols (ZKPs for privacy, MPC for confidential computation, blockchain for immutable recording), approximates the God Protocol without creating a single point of control. No individual agent possesses all the properties Szabo described. The network, collectively, approaches them.

This is the architecture that programmable trust infrastructure is building toward. The verification layer uses ZKPs and attestation protocols to prove facts without revealing data. The computation layer uses smart contracts, MPC, and (increasingly) FHE to perform computations on private inputs. The data access layer uses decentralized oracles like Chainlink and The Graph to bring real-world data on-chain with cryptographic guarantees. The consensus layer uses proof-of-stake validators and emerging data availability layers to maintain immutable, tamper-proof records.

The tradeoff is real. A distributed God Protocol is slower, less capable, and more complex than a centralized AGI would be. It cannot interpret ambiguity, assess intent, or navigate cultural context. It can only verify what is mathematically provable and execute what is algorithmically expressible. But it has a property that no centralized system can guarantee: no single entity controls it. In a world where the alignment problem remains unsolved and the governance of powerful AI systems is a subject of active political conflict, this property may be the most important one.

The useful path mixes centralized AGI with distributed cryptographic protocols. AI agents operating within cryptographic verification frameworks, where their outputs are constrained by zero-knowledge proofs, their data access is governed by MPC protocols, and their actions are recorded on immutable ledgers. This hybrid architecture captures some of the flexibility and contextual intelligence of AI while maintaining the trustlessness and transparency of cryptographic systems.

The Theological Endgame

Szabo's thought experiment, written as a two-page essay in 1997, has become the organizing framework for two of the most important technological movements of the twenty-first century: blockchain and artificial intelligence. Both are attempts to solve the same problem: how do you coordinate human activity without requiring trust in a fallible intermediary?

Blockchain approaches the problem from the bottom up, replacing specific TTPs with cryptographic protocols, one function at a time. Value transfer (Bitcoin). Computation (Ethereum). Privacy (ZKPs). Data feeds (oracles). Governance (DAOs). Each layer adds capability but remains constrained by the limits of deterministic code.

AGI approaches the problem from the top down, building a single entity that could, in principle, serve as a universal TTP for any domain. The capability is greater, but so is the risk. A universal TTP that is misaligned, captured, or corrupted is an existential threat, not a failed experiment.

The God Protocol remains an asymptote. A limit that can be approached but never reached. Every approximation introduces new tradeoffs: transparency vs. confidentiality, decentralization vs. capability, code vs. intent. The history of the twenty-first century may be written as the story of how closely humanity approached this limit and what was sacrificed along the way.