Author: a16z
Compiled by: Jiahuan, ChainCatcher
AI systems are disrupting an internet originally designed for human scale, as they make collaboration, transactions, and the generation of speech, video, and text unprecedentedly cheap, and this generated content is increasingly difficult to distinguish from human activity. We are already surrounded by CAPTCHAs; and now, we are beginning to see agents interacting and transacting like humans (as we reported here).
The problem is not the existence of AI; it's the internet's lack of a native way to distinguish between humans and machines while protecting privacy and usability.
This is where blockchain comes in. The arguments about how encryption can help build better AI systems (and vice versa) can be nuanced; therefore, we summarize in this article several reasons why AI needs blockchain more than ever.
Increasing the Cost of AI Impersonation
AI can forge voices, faces, writing styles, videos, and complete social personas on a large scale: an actor can, at an increasingly lower cost, masquerade as thousands of accounts, opinions, customers, or voters.
These impersonation strategies are not new. Any enterprising scammer could always hire voice actors, fake phone calls, or send phishing texts. What is new is the price: implementing these attacks on a large scale is becoming increasingly affordable.
At the same time, most online services assume one account corresponds to one person. When this assumption fails, everything downstream collapses. Detection-based methods (like CAPTCHAs) are bound to fail because AI advances faster than the tests designed to catch it.
So where does blockchain fit in? Decentralized "proof-of-human" or "proof-of-personhood" systems make it easy for one person to participate, but persistently difficult to impersonate many. While scanning an iris and obtaining a World ID might be relatively easy and affordable, obtaining a second one is almost impossible.
This makes it harder for AI to achieve large-scale impersonation by limiting the supply of IDs and increasing the marginal cost for attackers.
AI can forge content, but encryption makes low-cost forgery of human uniqueness extremely difficult. By restoring scarcity at the identity layer, blockchain increases the marginal cost of impersonation without adding friction to normal human behavior.
Creating Decentralized Proof-of-Personhood Systems
One way to prove you are human is through a digital ID, which contains everything needed to verify identity—username, PIN, password, and third-party proofs (e.g., citizenship or creditworthiness) and other credentials.
What does encryption add? Decentralization. Any identity system located at the center of the internet becomes a single point of failure. When agents act on behalf of humans—transacting, communicating, and coordinating—whoever controls the identity effectively controls the right to participate. Issuers can revoke access, impose fees, or assist in surveillance.
Decentralization reverses this dynamic: users, not platform gatekeepers, control their own identities, making them more secure and censorship-resistant.
Unlike traditional identity systems, decentralized proof-of-human mechanisms allow users to control and custody their own identities and verify their human status in a privacy-preserving and credibly neutral manner.
Creating Portable Universal "Passports" for Agents
AI agents do not reside in one place. A single agent might appear in chat apps, email exchanges, phone calls, browser sessions, and APIs. However, there is currently no reliable way to know that interactions in these different contexts point to the same agent, with the same state, capabilities, and authorizations provided by its "owner".
Furthermore, binding an agent's identity solely to one platform or marketplace renders it unusable in other products and all other important places.
A blockchain-based identity layer allows agents to have portable universal "passports". These identities can carry references to capabilities, permissions, and payment endpoints, and can be resolved anywhere, making agents harder to forge. This will also allow builders to create more useful agents and better user experiences: agents can exist in multiple ecosystems without worrying about being locked into any specific platform.
Enabling Machine-Scale Payments
As AI agents increasingly transact on behalf of humans, existing payment systems become a bottleneck. Large-scale agent payments will require new infrastructure, such as micropayment systems capable of handling tiny transactions across multiple sources.
Many existing blockchain tools—Rollups and L2s, AI-native financial institutions, and financial infrastructure protocols—show potential to solve this problem, enabling near-zero-cost transactions and more granular payment splits.
Crucially, these rails support machine-scale transactions that traditional financial systems cannot handle—micropayments, high-frequency interactions, and agent-to-agent commerce.
Nanopayments can be split among multiple data providers, allowing a single user interaction to trigger tiny payments to all contributing sources via automated smart contracts.
Smart contracts allow for executable retrospective payments triggered by completed transactions, compensating those information sources that contributed to the purchase decision in a fully transparent and traceable manner after the transaction occurs.
Blockchain enables the distribution of complex and programmable payment splits, ensuring revenue is fairly distributed through code-enforced rules rather than centralized decisions, thus establishing trustless financial relationships between autonomous agents.
Enforcing Privacy in AI Systems
At the core of many security systems lies a paradox: the more data they collect to protect users (e.g., social graphs, biometrics), the easier it becomes for AI to impersonate users.
This is where privacy and security become the same issue. The challenge is to make "proof-of-personhood" systems private by default and obscure information at every turn to ensure that only humans can generate the information needed to prove they are human.
Blockchain-based systems combined with zero-knowledge proof technology allow users to prove specific facts—PIN codes, ID numbers, eligibility criteria (e.g., drinking age for a bar)—without revealing the underlying data (e.g., address on a driver's license).
Applications gain the assurances they need, and AI systems are deprived of the raw materials needed for imitation. Privacy is no longer a feature layered on top; it is a core defense.
AI brings cheap scale but makes trust precarious. Blockchain successfully rebuilds trust by raising the cost of impersonation, guarding human-scale interactions, decentralizing identity, enforcing privacy by default, and endowing agents with native economic constraints.
If we desire an internet where AI agents can operate efficiently without eroding trust, blockchain is not an optional facility: it is the crucial piece that enables an AI-native internet to function well.
