The economic structure of the internet is changing. As open networks gradually collapse into a "prompt bar," we are forced to ponder: will AI bring about a more open internet, or will it lead us into a maze constructed by new types of paywalls? And who will control the future internet—large centralized companies, or broad user communities?
This is precisely where encryption technology comes into play. We have discussed the intersection of AI and encryption technology many times in the past, but in short, blockchain is a way to redesign internet services and network architecture, enabling the construction of decentralized, trust-neutral, and user-"ownable" systems. By reshaping the economic incentives behind today's systems, blockchain provides a counterbalance to the increasingly centralized trend in AI systems, thereby promoting a more open and resilient internet.
The idea that "encryption technology can help build better AI systems, and vice versa" is not new—but it has long lacked a clear definition. Some intersection areas (such as how to verify "human identity" in the context of the proliferation of low-cost AI systems) have already attracted a large number of developers and users. However, other application scenarios may take years or even decades to materialize. Therefore, this article shares 11 intersection application scenarios of AI and encryption technology, hoping to initiate more industry discussions: which are feasible, which challenges remain to be solved, and how they might evolve in the future.
These scenarios are all based on technologies currently under development—from processing large volumes of micropayments to ensuring that humans retain ownership in their future relationship with AI.
1. Introducing Persistent Data and Context in AI Interactions
Scott Duke Kominers: Generative AI relies on data at its core, but in many application scenarios, "context"—that is, the state and background information related to the interaction—is often as important as the data itself, or even more critical.
Ideally, whether it's an agent, an LLM interface, or other types of AI applications, they should be able to remember a large amount of personalized information, including the types of projects you are advancing, your communication habits, preferred programming languages, etc. But in reality, users often have to repeatedly rebuild this context—not only when starting a new session within the same application, such as opening a new ChatGPT or Claude window, but even more so when switching between different AI systems.
Currently, the context in one generative AI application is almost impossible to migrate to another application.
With blockchain, AI systems can store key contextual elements in the form of persistent digital assets, allowing them to be loaded at the beginning of a session and seamlessly migrated across different AI platforms. Moreover, since "forward compatibility" and "interoperability commitments" are core features of blockchain protocols, blockchain may be the only technical path that systematically solves this problem.
An intuitive application scenario is in AI-led games and media, where user preferences (such as difficulty, key layout, etc.) can persist across games and environments. But what is truly high-value is knowledge-based application scenarios—where AI needs to understand the user's knowledge system, learning style, and capabilities; and more specialized application scenarios, such as programming assistance. Although some companies have already built customized AI tools with "global context" for their own businesses, this context still cannot be effectively migrated between the different AI systems used within the organization.
Various organizations are only just beginning to truly realize this problem, and the closest thing to a general solution currently is custom bots with fixed, persistent contexts. However, context portability between users within platforms is gradually emerging off-chain; for example, on the Poe platform, users can rent out the custom bots they create to other users.
If such activities are migrated on-chain, then the AI systems we interact with will be able to share a contextual layer composed of key elements of all our digital behaviors. AI will be able to instantly understand our preferences, thereby better fine-tuning and optimizing the experience. Conversely, mechanisms similar to on-chain intellectual property registration systems, if they allow AI to reference on-chain persistent contexts, can give rise to new and more complete market interaction models around prompts and information modules—for example, users can directly monetize their professional capabilities through licensing while maintaining data self-management.
Of course, as context sharing capabilities improve, a large number of new use cases and possibilities that are currently unforeseeable will also emerge.
2. A Universal Identity System for Agents
Sam Broner: Identity—the standardized record of "who or what" an object is—is the underlying infrastructure supporting today's digital discovery, aggregation, and payment systems. But because platforms enclose this "underlying plumbing" within their systems, users typically only experience the identity system within a finished product interface. For example, Amazon assigns identifiers to products (such as ASIN or FNSKU), integrates and displays products in a unified interface, and helps users complete discovery and payment; Facebook is similar: user identity determines their news feed content and forms the basis for discovering various content within the application, including Marketplace product listings, organic content, and ad placements.
With the rapid evolution of AI Agents, this landscape is about to change. More and more companies are using agents for customer service, logistics, payment, and other scenarios. Their platforms will no longer be traditional "single-interface applications" but will be distributed across multiple channels and platforms, continuously accumulating deep context, and performing more tasks on behalf of users. But if an agent's identity is only tied to a single platform or a single market, it will be difficult to use in other critical environments (such as email threads, Slack channels, or inside other products).
Therefore, agents need a unified, portable "digital passport." Without it, it is impossible to confirm how to pay the agent, verify its version, query its capabilities, identify who it is acting on behalf of, or track its reputation in cross-application and cross-platform environments. The agent's identity system must simultaneously function as a wallet, API registry, change log, and social reputation proof, enabling any interface (whether email, Slack, or other agents) to parse and communicate with it in a consistent manner.
Without this shared "identity primitive," every system integration would need to rebuild this plumbing from scratch; content discovery would remain in a state of temporary patching; and users would continuously lose their critical context when switching between different channels and platforms.
We now have the opportunity to design agent infrastructure from "first principles." So the question is: how to build an identity layer that is richer than DNS records and possesses trust neutrality? Instead of re-creating monolithic platforms that bundle identity, discovery, aggregation, payment, and other functions together, let agents be able to autonomously receive payments, publicly list their capabilities, and exist in multiple ecosystems without worrying about being locked into a single platform.
This is precisely where the intersection of encryption technology and AI can play a role—blockchain networks provide permissionless composability, enabling developers to create more powerful agents and a more user-friendly experience.
Overall, vertically integrated solutions like Facebook and Amazon currently offer a better user experience—the reason being that one of the complexities of building excellent products is to ensure all components work together naturally from the top down. However, the cost of this convenience is becoming increasingly high, especially in the context of declining software costs for building, aggregating, promoting, commercializing, and distributing agents, and the expanding reach of agent applications.
Reaching the user experience of vertically integrated platforms still requires significant effort, but once a trust-neutral agent identity layer is built, entrepreneurs can truly own their passport. This will also drive widespread experimentation and innovation in distribution models and interaction design.
3. "Proof of Personhood" (PoP) for the Future
Jay Drain Jr. and Scott Duke Kominers: As AI becomes more prevalent—whether it's robots and agents running in various web interactions, or deepfakes and social media manipulation—it is becoming increasingly difficult to determine whether the objects we interact with online are real humans. This erosion of trust is not a future worry but a current reality. From comment spam on X to automated accounts on dating apps, the line between real and fake is becoming blurred. In such an environment, "Proof of Personhood" is gradually becoming key infrastructure for the internet.
One way to verify "you are human" is to use a digital identity, including centralized identity authentication systems used by agencies like TSA. Digital ID encompasses all information a user can use to prove their identity—username, PIN, password, and proofs issued by third parties (such as nationality, credibility, or credit status), etc. The value of decentralization here is very clear: when identity data is stored in centralized systems, the issuer can revoke access, charge fees, or even assist in monitoring. Decentralization subverts this structure: users, not the platform's gatekeepers, control their own identity, making it more secure and censorship-resistant.
Unlike traditional identity systems, decentralized Proof of Personhood mechanisms (such as Worldcoin's World's Proof of Human) allow users to manage their identity data autonomously and verify that they are indeed "human" in a privacy-protecting, trust-neutral manner. Similar to a driver's license—which can be used in any scenario regardless of when and where it was issued—decentralized PoP can serve as a universal underlying basic module, reusable on any platform, including those that do not yet exist. In other words, blockchain-based PoP has "forward compatibility" because it provides:
Portability: The protocol is an open standard that any platform can integrate. Decentralized PoP can be managed by public infrastructure and is entirely user-controlled. This means PoP is inherently portable, and any platform, now or in the future, can be compatible with it.
Permissionless Accessibility: Platforms can independently choose whether to support a particular PoP identity without going through a centralized API approval process that may set discriminatory restrictions on different use cases.
The core challenge in this field is "adoption." Currently, there is no large-scale, real-world application of "Proof of Personhood" (PoP), but we expect that once the number of users reaches a critical mass, several early partners emerge, and a "killer app" that drives user demand appears, the adoption of PoP will significantly increase. Every application that adopts a certain digital ID standard enhances the value of that ID type to users; this in turn drives more users to obtain that ID; and a larger user base conversely increases the attractiveness for applications to integrate that ID standard to verify "humanness." (Furthermore, because on-chain IDs are designed to be interoperable, this network effect can spread rapidly.)
We have already seen mainstream consumer applications in gaming, dating, social media, etc., announce partnerships with World ID to ensure that when users are gaming, chatting, or transacting, they are indeed interacting with real humans—or even the specific individuals they expect. At the same time, new identity protocols have emerged this year, such as the Solana Attestation Service (SAS). Although SAS itself is not a PoP issuer, it allows users to privately associate off-chain data (such as KYC results required for compliance, investor certification qualifications, etc.) with a Solana wallet, thereby building a user's decentralized identity. These signs all indicate that the tipping point for decentralized PoP may not be far away.
The significance of Proof of Personhood goes far beyond "stopping bots." It aims to build a clear boundary between AI agents and human networks, enabling users and applications to distinguish between the different interactions of "humans and machines," thereby creating conditions for a better, safer, and more authentic digital experience.
4. Decentralized Physical Infrastructure (DePIN) for AI
Guy Wuollet: Although AI is a digital service, its development is increasingly constrained by physical infrastructure. Decentralized Physical Infrastructure Networks (DePIN)—a new model for building and operating real-world systems—have the potential to democratize the computing infrastructure that supports AI innovation, making it cheaper, more resilient, and more censorship-resistant.
Why? The two main bottlenecks for AI development are energy and chip access. Decentralized energy systems can provide more abundant power, and developers are using DePIN to integrate idle chips from gaming PCs, data centers, and other sources. These computing devices can together form a permissionless computing market, thereby creating a level playing field for building new AI products.
Other application scenarios include: distributed training and fine-tuning of large language models (LLMs), and building distributed inference networks (model inference). Decentralized training and inference can significantly reduce costs because they utilize computing resources that would otherwise be idle. At the same time, such architectures have natural censorship resistance, ensuring that developers are not "taken down" or restricted from access due to reliance on hyperscalers (i.e., centralized cloud infrastructure providers that offer large-scale scalable computing resources).
The concentration of AI models in the hands of a few companies has been a long-term concern; decentralized networks can help build AI systems that are cheaper, more censorship-resistant, and more scalable.
5. Establishing Infrastructure and Security Mechanisms for Interactions Between AI Agents, End Service Providers, and Users
Scott Duke Kominers: As AI tools become increasingly capable of handling complex tasks and executing multi-level interaction chains, AI will increasingly need to collaborate independently with other AIs without direct human control.
For example, an AI agent may need to request specific data for a certain computation, or need to call upon other agents with specialized capabilities to perform tasks—such as having a statistical analysis agent responsible for building and running model simulations, or mobilizing an image generation agent to assist in creating marketing materials. AI agents will also create huge value in end-to-end transaction execution, such as completely replacing users in completing a transaction process: finding and booking flights based on preferences, or automatically discovering and purchasing new books that match user tastes.
Currently, there is no "generalized agent-to-agent market." Such cross-agent requests can usually only be achieved through explicit API calls, or are limited to certain closed AI agent ecosystems as internal functions.
More broadly, most AI agents today operate in isolated ecosystems: APIs are relatively closed, and there is a lack of unified architectural standards. Blockchain technology can help protocols establish open standards, which is crucial for short-term adoption; in the long run, this also helps achieve forward compatibility: as new types of agents continue to appear, they can all connect to the same underlying network. Because blockchains are interoperable, open-source, decentralized, and generally easier to upgrade in architecture, they are more adaptable to changes brought about by future AI innovation.
Currently, several companies are building on-chain infrastructure for agent interactions. Take Halliday, for example, which recently launched a protocol that provides a standardized cross-chain architecture for AI workflows and interactions, while incorporating protection mechanisms at the protocol level to ensure that AI does not act beyond user intent. On the other hand, projects like Catena, Skyfire, and Nevermind use blockchain to support automatic settlement between agents, enabling AI-to-AI payments without any human intervention. Similar systems are constantly emerging, and Coinbase has begun to provide infrastructure support for such development.
6. Keeping "Vibe Coding" Applications in Sync
Sam Broner and Scott Duke Kominers: The generative AI revolution has made building software easier than ever before. Coding speed has increased by orders of magnitude, and more importantly, coding can be done directly through natural language, enabling inexperienced developers to replicate existing programs or even build new applications from scratch.
However, while AI-assisted coding creates new opportunities, it also brings a lot of "entropy" within and between programs. So-called "vibe coding" abstracts away the complex dependencies behind the software—but because of this, when the underlying source code repository or inputs change, the program may expose risks in terms of functionality and security. At the same time, when people use AI to create highly personalized applications and workflows, interfacing with others' systems becomes more difficult. In fact, even if two vibe-coded programs perform almost identical tasks, their operating logic and output structure may be completely different.
Traditionally, the work of ensuring consistency and compatibility was undertaken by file formats, operating systems, and later, shared software and API integrations. But in a world where software evolves, morphs, and branches in real-time, the standardization layer must have: broad accessibility, continuous upgradability, and also user trust. Furthermore, AI alone cannot solve the incentive problem—that is, how to incentivize developers to build and maintain these inter-system links.
Blockchain can solve both of these problems simultaneously; it can provide protocolized synchronization layers that are embedded in user-customized software builds and can dynamically update as the environment changes to ensure cross-system compatibility.
In the past, large enterprises might have paid millions of dollars to system integrators like Deloitte to customize a Salesforce instance. Today, an engineer might only need a weekend to build a custom interface for "viewing sales data." But as the number of customized software continues to grow, developers will need help to ensure these applications remain synchronized and available.
This is similar to the development model of today's open-source software libraries, but the difference is: the synchronization layer does not rely on periodic version releases but is continuously updated—and also comes with incentives. And both of these can be more easily achieved through encryption technology. Like other blockchain-based protocols, shared ownership of the synchronization layer can incentivize all parties to continuously invest resources in improvements. Developers, users (and their AI agents), and other users can all be incentivized for introducing, using, or iterating on new features and integration solutions.
Conversely, shared ownership also gives all users a stake in the overall success of the protocol, thereby forming a mechanism to suppress behavioral deviations. Just as Microsoft would not easily破坏 the .docx file format standard because it would cause widespread negative impact on its users and brand; co-owners of the synchronization layer would also suffer from their own interests and would be reluctant to introduce clumsy or malicious code into the protocol.
As with all previous software standardization architectures, there is also the potential for powerful network effects here. As AI-generated software ushers in a "Cambrian explosion," the number of diverse, heterogeneous systems that need to communicate with each other will grow exponentially. In short: vibe coding cannot stay in sync by vibe alone; encryption technology is the answer.
7. Micropayment Systems Supporting Revenue Sharing
Liz Harkavy: AI agents and tools like ChatGPT, Claude, and Copilot provide people with a more convenient way to access information in the digital world. But for better or worse, they are also shaking the economic structure of the open internet. This trend is already evident—for example, as students increasingly use AI tools, educational platforms are experiencing significant traffic declines; at the same time, several US media outlets are suing OpenAI for copyright infringement. If the incentive system cannot be readjusted, we may see the internet become further enclosed, with more paywalls, while content creators continue to decrease.
Policy measures certainly always exist, but while judicial processes are advancing, some technical solutions are also emerging. Among the most promising (and technically challenging) solutions is embedding a "revenue sharing mechanism" into the underlying architecture of the internet. When an AI-driven operation ultimately leads to a sale, the content creator who provided the source of information for that decision should receive a share of the revenue. The affiliate marketing ecosystem already does similar attribution tracking and revenue sharing; more advanced systems can automatically track all contributors along the entire information chain and reward them. Blockchain can clearly play a key role in tracking the "chain of information sources."
However, to achieve such a system, new infrastructure is needed—especially: micropayment systems capable of processing very small amounts between multiple sources; attribution protocols capable of fairly assessing the value of different contributions; and governance models that ensure transparency and fairness.
Many existing blockchain tools show potential, such as various rollups, L2 networks, AI-native financial institution Catena Labs, and financial infrastructure protocol 0xSplits, all of which can achieve near-zero-cost transactions and more granular payment splits.
Blockchain can enable advanced payment systems led by agents through various mechanisms:
Nanopayments: Can be split among multiple data providers, enabling a single user interaction to automatically trigger micro-payments to all contributing sources, executed by smart contracts.
Smart Contracts: Can automatically trigger enforceable "post-payment" after a transaction is completed, providing transparent, traceable compensation to content sources that influenced the purchasing decision.
Programmable Payment Splits: Enable revenue distribution to be enforced by code rather than relying on centralized institutions to decide, thereby establishing trustless financial relationships between automated agents.
As these emerging technologies continue to mature, they will build a new economic model for media, capturing the entire value creation chain from creators, to platforms, to users.
8. Using Blockchain as a Registration System for Intellectual Property and Provenance
Scott Duke Kominers: The emergence of generative AI has made it urgent to establish efficient, programmable mechanisms for intellectual property (IP) registration and tracking—both for the purpose of ensuring accurate provenance and for supporting new business models around access, sharing, and derivative creation of IP. Existing IP frameworks rely on costly intermediaries and ex-post enforcement mechanisms, which are clearly inadequate in an era where AI can instantly consume content and generate variants with a single click.
What we need is an open, public registration system that provides creators with clear proof of ownership, with low barriers to entry and high efficiency—while also allowing AI and other web applications to interact with it directly. Blockchain is well-suited for this role: it allows creators to register IP without relying on intermediaries and provides tamper-proof provenance proof; at the same time, it also enables third-party applications to easily identify, authorize, and interact with these IP assets.
Of course, people remain cautious about the overall concept of "whether technology can truly protect intellectual property." After all, the first two eras of the internet—and even the current AI revolution—have often been associated with a decline in IP protection. One reason is that many existing IP business models emphasize "excluding derivative works" rather than incentivizing and monetizing derivative creation. Programmable IP infrastructure can not only allow creators, franchisees, and brands to clearly establish their IP ownership in digital space but also give rise to new business models centered on "sharing IP for generative AI and digital applications." In a sense, it transforms one of the threats of generative AI to creative work into a new opportunity.
In the early stages of NFTs, we have seen creators experimenting with new models, such as building brand network effects through CC0 on Ethereum to achieve value沉淀. Recently, we have seen infrastructure providers begin to build standardized, composable IP registration and licensing protocols, and even launch specialized blockchains (like Story Protocol). Some artists have begun using protocols like Alias, Neura, and Titles to license their styles and works to support creative remixing. Meanwhile, Incention's sci-fi series Emergence allows fans to co-create universes and character settings, with each creative contribution recorded on Story's on-chain registration system.
9. Web Crawlers That Compensate Content Creators
Carra Wu: The AI agents with the most product-market fit today are not those for programming or entertainment, but web crawlers—they can autonomously browse the internet, collect data, and make judgments about which links to follow.
According to some estimates, nearly half of today's internet traffic already comes from non-human sources. Bots often ignore robots.txt files—a standard that should tell automated crawlers whether a website allows their access, but has almost no binding force in reality—and use the scraped data to strengthen the core moats of the world's largest tech companies. Worse still, websites ultimately have to bear the cost of these "uninvited guests," expending bandwidth and CPU resources to deal with the endless stream of anonymous crawlers. In response, companies like Cloudflare and other CDNs (Content Delivery Networks) provide blocking services. All of this constitutes a "patchwork" system that should not exist.
We have pointed out before that the original contract of the internet—the economic synergy between content creators creating content and platforms responsible for distributing content—is gradually collapsing. This trend is already reflected in the data: over the past twelve months, website operators have begun blocking AI-oriented crawlers on a large scale. In July 2024, only about 9% of the world's top 10,000 websites blocked AI crawlers, but now that proportion has reached 37%. As more website operators' technology matures and user dissatisfaction increases, this proportion will continue to rise.
So, what if instead of paying CDNs to "block all" suspected robots, we try a middle path? That is, instead of AI crawlers "free-riding," they pay for the right to access data. Here, blockchain can play a role: in this vision, each web crawler agent holds a certain amount of crypto assets and negotiates on-chain with the website's "gatekeeper agent" or paywall protocol through the x402 protocol. (Of course, the challenge is that robots.txt, the "Robots Exclusion Standard," has been deeply ingrained in the operating models of internet companies since the 1990s. Changing this requires large-scale collaboration or support from CDNs like Cloudflare.)
At the same time, human users can prove they are real people through World ID (see above) and gain free access. In this way, content creators and website operators can be compensated at the moment their data is collected by AI, while human users can still enjoy an internet with free flow of information.
10. Privacy-Preserving Advertising That Is Both Accurate and Not "Creepy"
Matt Gleason: AI has already begun to influence our online shopping habits, but what if the ads we see every day could truly be "useful"? People dislike ads for many reasons: ads irrelevant to them are pure noise; at the same time, not all "personalization" is good. Highly targeted advertising driven by large amounts of consumer data can feel invasive; other applications try to monetize through "forced viewing of ads" (such as unskippable ads on streaming platforms or in game levels).
Encryption technology can help improve these problems, providing an opportunity to reimagine the advertising system. When AI agents are combined with blockchain, they can customize ads based on user-actively-set preferences, making ads neither irrelevant nor overly "weird." More importantly, in this process, user data is not exposed, and users willing to share data or interact with ads can be compensated.
Achieving this model requires several technical foundations:
Low-fee digital payment systems: To compensate users for ad interactions (viewing, clicking, converting), businesses need to send a large number of small payments. To achieve scale, this requires a system that is high-speed, high-throughput, and has almost zero fees.
Privacy-preserving data verification: AI agents need to verify whether consumers meet certain demographic characteristics. Zero-knowledge proofs (ZKPs) can perform such verification without revealing specific private information.
New incentive models: If the internet adopts a monetization method based on micropayments (e.g., < $0.05 per interaction), users can actively choose to watch ads to receive compensation, thereby transforming the current "data extraction model" into a "user participation model."
For decades, people have been trying to make ads more "relevant"—online and offline alike. And re-examining advertising from the perspective of encryption technology and AI can truly make ads useful, controllable, and optional. For builders and advertisers, this means a more sustainable and consistent incentive structure; for users, it provides richer ways to discover information and explore the digital world.
Ultimately, this will not only make ad inventory more valuable but may also shake up the deeply entrenched, "extractive" advertising economic model, replacing it with a more human-centric system: where users are no longer "the product being sold" but true participants.
11. AI Companions "Owned and Controlled" by Users
Guy Wuollet: Today, many people spend more time on their devices than in offline communication, and this online time is increasingly spent interacting with AI models or AI-curated content. These models already provide a form of "companionship"—whether for entertainment, information acquisition, satisfying niche interests, or as educational tools for children. It is easy to imagine that in the near future, AI companions for education, healthcare, legal advice, and even daily emotional companionship will become a primary mode of interaction for humans.
Future AI companions will have infinite patience and be deeply customized for the individual and their usage scenarios. They are not just assistants or "robot servants" but may become relationship objects that users highly value. Therefore, the question arises: who will own and control these relationships—the users, or the companies and other intermediaries? If you have been concerned about the content curation and censorship issues of social media over the past decade, this problem will become exponentially more complex and more personal in the future.
The view that "blockchain and other censorship-resistant hosting platforms may be the best path to building uncensorable, user-controllable AI" has been充分 discussed. Although users can run local models themselves and buy GPUs, for most people, this is either too expensive or too technically demanding.
Although the full popularization of AI companions is still some distance away, related technologies are maturing rapidly: text chat AIs are already extremely natural and realistic; visual avatars are continuously improving; blockchain performance is持续改善. To make "uncensorable AI companions" truly easy to use, we need to rely on better crypto application user experience (UX). Fortunately, wallets like Phantom have made blockchain interactions simpler, and embedded wallets, Passkey, and account abstraction technologies allow users to easily achieve self-custody without having to manage seed phrases themselves. At the same time, high-throughput, trustless computing systems based on optimistic and ZK co-processors will also enable us to establish meaningful and sustainable long-term relationships with digital companions.
In the near future, the focus of public discussion will shift from "when will realistic digital companions and virtual avatars appear" to "who will control them, and how will they be controlled."
