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.

Artificial intelligence significantly reduces the costs of scaled operations but makes trust difficult to establish, whereas blockchain technology can reshape the trust system.

Article by: a16z

Translation: Chopper, Foresight News

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

The key issue is not the existence of AI 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 underpinned by deep logical reasoning. 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, enabling large-scale operations: a single agent can impersonate thousands of accounts, simulating different viewpoints, consumers, or voters, with the cost of such operations continually decreasing.

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

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

So how can blockchain help? Decentralized human proof or identity systems allow users to easily complete a single identity verification, fundamentally preventing multiple identities for one person. For example, scanning an iris to obtain a global identity identifier may be simple and economical, 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 AI impersonation operations difficult to execute.

AI can forge content, but cryptographic technology prevents it from cheaply forging unique human identities. Blockchain reshapes scarcity at the identity layer, raising the marginal costs of impersonation without adding extra barriers to normal human use.

Building a Decentralized Human Identity System

One way to prove human identity is through digital identity identifiers, 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 platform gatekeepers, control their own identity information. This makes identity identifiers 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 neutral manner.

Creating Portable, Universal “Digital Passports” for AI Agents

AI agents are not tied to a single platform: an 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 owner authorization.

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

A blockchain-based identity layer can create portable, universal “digital passports” for AI agents. These identifiers 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 Scaled Payment Transactions

As AI agents increasingly transact on behalf of humans, existing payment systems become obvious bottlenecks. Large-scale AI payments require new infrastructure capable of handling microtransactions from multiple sources.

Many blockchain-based tools—such as roll-up solutions, second-layer networks, native AI financial institutions, and financial infrastructure protocols—show potential to solve this problem, enabling 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 traditional finance cannot handle.

• Ultra-small payments can be split among multiple data providers, with automated smart contracts triggering small payments to all relevant data sources during a single user interaction;
• Smart contracts support traceable, enforceable payments based on completed transactions, compensating entities that provide information for purchasing decisions, with full transparency and traceability;
• Blockchain enables complex, programmable payment splits, with code-enforced rules ensuring fair revenue distribution, rather than relying on centralized decision-making, establishing trustless financial relationships among autonomous 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 user identities.

In this context, privacy protection and security become intertwined. The challenge is to design human identity systems with default 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 prove specific facts—such as personal ID numbers, citizenship, or qualification standards—without revealing underlying raw data (e.g., address on a driver’s license).

This way, application providers gain the necessary identity verification, while AI systems are deprived of the raw data needed for impersonation. Privacy protection is no longer an add-on but a core defense against AI impersonation.

Summary

AI significantly lowers the costs of scaled operations but makes trust difficult to establish. Blockchain technology can reshape the trust system: increasing the cost of impersonation, maintaining human-scale interaction modes, enabling decentralized identity systems, protecting privacy by 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, then blockchain is not optional. It is the essential underlying technology for creating an AI-native internet—an integral part currently missing from the existing web.