a16z: Why does AI urgently need cryptography?
As artificial intelligence continues to advance rapidly, the importance of securing AI systems and data becomes increasingly critical. Cryptography provides essential tools to protect sensitive information, ensure data integrity, and maintain privacy in AI applications. From safeguarding training datasets to securing communication channels between AI components, encryption techniques help prevent malicious attacks and unauthorized access. As AI integrates more deeply into our daily lives, the role of cryptography in building trustworthy and secure AI systems is more vital than ever.

Written by: a16z

Translated by: Chopper, Foresight News

Artificial intelligence systems are disrupting the internet originally designed on a human scale. They reduce collaboration and transaction costs to historic lows, and the generated speech, video, and text are increasingly indistinguishable from human behavior. Today, we are already troubled by human-machine verification, and currently, AI agents are beginning to interact and transact like humans.

The key issue is not the existence of artificial intelligence, but the lack of a native mechanism on the internet that can distinguish humans from machines while protecting privacy and ensuring ease of use.

This is where blockchain technology comes into play. Cryptography helps create better AI systems, and conversely, AI can empower cryptographic techniques—both backed by deep underlying logic. Here, we summarize several reasons why AI now needs blockchain more than ever.

Increasing the Cost of AI Impersonation

AI can forge voices, facial features, writing styles, video content, and even create complete social personas, capable of large-scale operations: a single AI agent can impersonate thousands of accounts, simulate different viewpoints, consumers, or voters, and the cost of such operations continues to decrease.

These types of impersonation are not new: malicious actors have long been able to hire voice actors, forge phone calls, and send phishing messages. The real change is in the cost: today, the threshold for large-scale fraudulent attacks has significantly lowered.

Meanwhile, most online services assume “one account per real user.” When this premise fails, all subsequent systems collapse. Detection-based countermeasures (like human-machine verification) are ultimately doomed to fail because AI evolves much faster than detection techniques designed specifically for it.

So, how can blockchain play a role? Decentralized human verification or identity systems allow users to easily complete a single identity verification, fundamentally preventing one person from having multiple identities. For example, scanning an iris to obtain a global identity identifier may be simple and cost-effective, but obtaining a second identifier is nearly impossible.

By limiting the issuance of identity credentials and increasing the marginal costs for attackers, blockchain makes large-scale impersonation by AI difficult to achieve.

AI can forge content, but cryptography prevents it from cheaply forging unique human identities. Blockchain reshapes scarcity at the identity layer, raising the marginal costs of impersonation while not adding extra barriers to normal human use.

Building a Decentralized Human Identity System

One way to prove human identity is through digital identity credentials, which encompass all information used for verification: usernames, personal identification numbers, passwords, third-party attestations (such as citizenship or creditworthiness), and other relevant credentials.

What value does cryptography add? The answer is decentralization. Any centralized identity system at the core of the internet can become a single point of failure. When AI agents act on behalf of humans for transactions, communication, and collaboration, control over identity verification effectively grants participation rights. Centralized issuers can revoke user permissions at will, charge fees, or even facilitate surveillance.

Decentralization reverses this pattern: users, not platforms, control their own identity information. This makes identity credentials more secure and resistant to censorship.

Unlike traditional identity systems, decentralized human proof mechanisms allow users to autonomously control and store their identity data, completing human verification in a privacy-preserving and fully neutral manner.

Creating a Portable, Universal “Digital Pass” for AI Agents

AI agents are not tied to a single platform: one agent can appear across various chat apps, email conversations, phone calls, browser sessions, and APIs. Currently, there is no reliable mechanism to confirm that interactions across these different scenarios originate from the same AI agent with consistent state, capabilities, and authorized by its “owner.”

Furthermore, if an AI agent’s identity is only linked to one platform or marketplace, it cannot be used in other products or critical scenarios. This results in fragmented user experiences and cumbersome, inefficient scene-specific adaptations.

A blockchain-based identity layer can create a portable, universal “digital pass” for AI agents. These identity credentials can link to the agent’s capabilities, permissions, and payment endpoints, and can be verified in any scenario, greatly increasing the difficulty of impersonating AI agents. This also enables developers to build more practical AI agents, providing better user experiences: agents can operate across multiple ecosystems without being bound to a specific platform.

Enabling Scalable Payments

As AI agents increasingly transact on behalf of humans, existing payment systems become a clear bottleneck. Large-scale AI payments require new infrastructure capable of handling numerous small transactions—micro-payments.

Many blockchain-based tools, such as roll-up solutions, second-layer networks, AI-native financial institutions, and financial infrastructure protocols, show potential for solving this problem. They can facilitate near-zero-cost transactions and more granular payment splitting.

The key is that these blockchain payment infrastructures can support machine-scale transactions, including micro-payments, high-frequency interactions, and commercial exchanges between agents—things that traditional financial systems cannot handle.

Ultra-small payments can be split among multiple data providers. Automated smart contracts can trigger small payments to all relevant data providers with each user interaction.

Smart contracts support traceable, enforceable payments based on completed transactions. After a transaction, compensation can be provided to entities that contributed information for decision-making, with full transparency and traceability.

Blockchain enables complex, programmable payment splits, ensuring fair revenue distribution through code-enforced rules, rather than relying on centralized decision-making. This fosters trustless financial relationships among autonomous AI agents.

Protecting Privacy and Security in AI Systems

Many security systems face a paradox: the more data collected to protect users, the easier it becomes for AI to impersonate users.

In this context, privacy protection and security are intertwined. The challenge is to design human identity systems with built-in privacy features, hiding sensitive information at all stages, ensuring that only genuine humans can provide the necessary proof of identity.

Combining blockchain systems with zero-knowledge proof technology allows users to verify specific facts—such as personal ID numbers, citizenship, or qualifications—without revealing underlying raw data (e.g., address on a driver’s license).

This way, applications can obtain the required identity verification, while AI systems are deprived of the raw data needed for impersonation. Privacy protection becomes not just an add-on, but the core defense against AI impersonation.

Summary

AI significantly reduces the costs of large-scale operations but makes trust harder to establish. Blockchain technology can reshape the trust system: increasing the costs of impersonation, maintaining human-scale interaction models, enabling decentralized identity systems, making privacy protection the default, and providing native economic constraints for AI agents.

If we want to build an internet where AI agents can operate normally without destroying trust, blockchain is not optional. It is the essential underlying technology for constructing an AI-native internet—an indispensable core component currently missing from the existing internet.