🧠 Theme 1: Decentralized AI & Intelligence Markets
The core thesis: AI is a commodity. It should be produced, priced, and traded in open markets — not siloed inside closed corporate labs.
Bittensor: A Peer-to-Peer Intelligence Market
Jacob Robert Steeves & Ala Shaabana · 2021
The foundational paper for decentralized ML. Bittensor proposes a blockchain network where machine learning models compete peer-to-peer and are rewarded with TAO tokens based on their value to other models — not human judges. If you’re building anything in the decentralized AI compute or model marketplace space, this is the paper your architecture is implicitly responding to. Grayscale filed the first US spot ETF for TAO in December 2025.
SingularityNET: A Decentralized, Open Market for AI Services
Ben Goertzel et al. · 2017
One of the earliest and most ambitious papers in this space. Proposes a blockchain-based marketplace where AI services are bought and sold autonomously by both humans and AI agents. Still highly relevant as a conceptual framework for AI-to-AI economic transactions — now one of the central ideas in the agentic AI conversation of 2025–26.
Ocean Protocol: A Decentralized Substrate for AI Data Exchange
Trent McConaghy et al. · 2019
Data is the raw material of AI. Ocean Protocol builds a privacy-preserving marketplace for data using blockchain. Its “compute-to-data” architecture — where the AI goes to the data rather than the data going to the AI — is directly practical for founders building AI products that require proprietary or sensitive datasets.
Fetch.ai: The Economic Internet
Humayun Sheikh, Jonathan Ward et al. · 2019
Introduced the concept of “autonomous economic agents” — software agents that can represent individuals, organizations, or devices, discover each other, negotiate, and transact without human intervention. Now merged into the Artificial Superintelligence Alliance (ASI), this whitepaper is the conceptual backbone of the decentralized agent economy every AI × DeFi builder is working toward.
🤖 Theme 2: AI Agents, Identity & Accountability
The hardest unsolved problem in agentic AI: how do you know who an agent is, who sent it, and who is responsible for what it does?
The Agent Economy: A Blockchain-Based Foundation for Autonomous AI Agents
Arxiv · February 2026
The most comprehensive recent treatment of what an “agent economy” actually means structurally. Covers decentralized identity for agents, cryptographic wallets, on-chain reputation, and AI agents as genuine economic peers to humans — not just tools. Six core research challenges are identified. If you’re building agent infrastructure or agent-to-agent transaction systems, this is the current state-of-the-art map.
Identity Management for Agentic AI
Tobin South (Lead Editor) · MIT · October 2025
A practical whitepaper on the authentication, authorization, and identity challenges that arise when AI agents act autonomously. Covers MCP (Model Context Protocol), delegated authority, agent-centric identities, and scalable access control. Essential reading for any founder whose product involves agents taking actions on behalf of users.
Know Your Agent: Governing AI Identity on the Agentic Web
Tomer Jordi Chaffer · 2024
Applies KYC logic to AI agents — asking what it would mean to verify, monitor, and hold accountable an AI agent operating on public networks. Uses Olas, Zerebro, and other live systems as case studies. Relevant for founders building compliance-aware agent platforms or anyone thinking about the governance layer above agent autonomy.
AI Agents with Decentralized Identifiers and Verifiable Credentials
Arxiv · November 2025
A prototypical framework where each AI agent is given a self-sovereign digital identity using W3C DIDs and Verifiable Credentials — allowing agents to authenticate each other and establish cross-domain trust without a central authority. If you’re building multi-agent systems or agent marketplaces, this is the identity architecture worth understanding.
Blockchain Infrastructure as a Trust Layer for AI Systems
Jung-Hua Liu · February 2026
A comprehensive synthesis of how blockchain can serve as the trust layer for AI — covering identity, payments, and privacy in a world where AI-generated content is indistinguishable from human-generated content. Strong on the economic framing: AI reduces the marginal cost of impersonation to near zero, and blockchain is one of the few mechanisms capable of restoring trust at scale.
🔐 Theme 3: Privacy, Verifiability & zkML
Can you prove an AI model ran correctly — without revealing the model or the data? This is the central technical challenge for trustless AI on-chain.
zkML: Zero-Knowledge Machine Learning
Modulus Labs · 2023
The paper that benchmarked ZK proof systems against ML architectures and made the tradeoffs concrete for the first time. If you’re building verifiable AI inference — where on-chain contracts need to trust AI outputs without seeing the model — this is the starting point. The “Cost of Intelligence” report from the same team is a strong companion piece.
A Survey of Zero-Knowledge Proof Based Verifiable Machine Learning
Springer Nature / Arxiv · 2025
The most up-to-date comprehensive survey of zkML research from 2017 to mid-2025. Covers algorithmic categories, technical evolution, implementation bottlenecks, and commercial deployment. Essential if you’re evaluating which ZK approach (zk-SNARKs, zk-STARKs, halo2, Plonky2) is right for your product’s compute and latency constraints.
Privacy-Preserving Decentralized AI with Confidential Computing
Dayeol Lee, Jorge António, Hisham Khan (Atoma Network) · October 2024
Argues that zkML isn’t yet production-ready for LLMs — the overhead is prohibitive — and proposes Confidential Computing via hardware Trusted Execution Environments (TEEs) as the practical near-term solution for privacy in decentralized AI inference. Honest about tradeoffs. Key reading for founders building private AI inference infrastructure.
Framework for End-to-End Verifiable AI Pipelines
Arxiv · March 2025
Extends verifiability from single inference steps to full AI pipelines — training, fine-tuning, retrieval, and inference. Introduces “provenance chains” for AI-generated decisions: when an agent makes a decision, you can prove what model ran, on what data, with what parameters. Directly relevant to the legal accountability gap.
A Scalable, Privacy-Preserving Decentralized Identity Framework
Arxiv · 2025
Combines W3C DIDs and Verifiable Credentials with zero-knowledge proofs to enable identity verification without revealing the underlying data. Directly applicable to founders building privacy-first onboarding, KYC-lite credential systems, or agent authentication layers.
💻 Theme 4: Data Markets, Compute & DePIN
AI needs three inputs: compute, data, and models. Decentralized infrastructure is being built for all three.
Render Network: Decentralized GPU Compute for AI and Creative Workloads
Jules Urbach · 2017 (updated 2023)
The original vision for decentralized GPU rendering, now expanded to AI compute workloads. With NVIDIA’s Blackwell architecture now onboarded by Render nodes, this is live infrastructure. For founders building AI products with high compute costs, understanding the Render model is directly relevant to your cost structure.
Numerai: A Decentralized Hedge Fund Built on Encrypted Data
Richard Craib · 2017
Numerai encrypts its financial data and distributes it to thousands of data scientists who build ML models without ever seeing the raw data. The best models are staked with NMR tokens and rewarded. It’s the first production-scale implementation of the “data goes to the model” architecture that Ocean Protocol and others later theorized. Still running after nearly a decade.
Worldcoin: A New Identity and Financial Network
Sam Altman, Alex Blania et al. · 2022
The most controversial paper in this list — and one of the most important. Worldcoin proposes biometric proof-of-personhood using iris scans as the foundation for distinguishing humans from AI agents at scale. The underlying problem it addresses — “how do you know if you’re talking to a human?” — is now one of the most urgent questions on the internet. Read it critically, but read it.
🏛️ Theme 5: Governance, DAOs & AI Agents
Who governs AI systems when there’s no CEO? What happens when DAOs deploy AI agents?
Decentralized Governance of Autonomous AI Agents
Arxiv · December 2024
Directly addresses the hardest governance question in the AI × crypto space: when a DAO deploys an autonomous AI agent that causes harm, who is responsible? Proposes governance frameworks for distributed accountability of AI agents operating in decentralized systems. Required reading if you’re building any DAO-controlled agent infrastructure.
How to Count AIs: Individuation and Liability for AI Agents
Arxiv · March 2026
A legal-philosophical paper with direct practical implications. As AI agents proliferate, how do we count them as distinct entities? How do liability, identity, and accountability attach to agents that can be copied, forked, or run as multiple simultaneous instances? Directly relevant to any founder designing agent architectures where legal standing or accountability could arise.
AI-Based Crypto Tokens: The Illusion of Decentralized AI?
Mafrur et al. · IET Blockchain · 2025
A critical perspective — included deliberately. Examines whether projects like Bittensor, Render, and Ocean Protocol are actually decentralized in practice, or whether they replicate centralization under a different label. Raises the DAO liability issue, token-weighted governance problems, and the gap between whitepaper claims and on-chain reality. Every founder building in this space should be able to answer the questions this paper raises.
📚 Foundational Papers Worth Revisiting
These aren’t AI × crypto papers — but they’re the bedrock that everything above builds on.
| bitcoin.org/bitcoin.pdf | | ethereum.org/whitepaper | | arxiv.org/abs/1706.03762 | | arxiv.org/pdf/1407.3561 | | eprint.iacr.org/2014/349.pdf | | eips.ethereum.org/eip-4337 | | w3.org/TR/did-core |
What’s Not on This List — And Why
GPT-4 Technical Report (OpenAI, 2023) — important for understanding frontier model capabilities, but not specific to the crypto intersection. Read it separately.
The Alignment Problem papers (RLHF, Constitutional AI, etc.) — critical for anyone building AI products, but covered better elsewhere. Start with Anthropic’s Constitutional AI paper.
Project-specific tokenomics papers — most are marketing dressed as research. The papers above are the ones with durable technical or conceptual value beyond their own ecosystem.