Cloudflare adopts x402 standard, opening Bitcoin micropayments for AI agents

Full text

Bitcoin Magazine

Cloudflare x402 Integration Opens Door for Bitcoin in AI Agent Micropayments

Cloudflare recently announced the launch of its monetization program via the Coinbase-led x402 machine payments standard. x402, which lets AI agents pay for data online with crypto, has been gaining steam among the AI-pilled, as it unlocks more capable agent interactions with the open web.

Cloudflare, founded in 2009, has grown from a DDoS mitigation and content delivery network (CDN) provider into one of the internet’s foundational infrastructure companies.

The company, which launched publicly in 2010, had the mission to make web performance and security accessible to everyone, not just large enterprises. Today, Cloudflare powers approximately 20-23% of all websites globally, handles tens of millions of HTTP requests per second across 330+ cities in over 100 countries , touching a significant portion of global internet traffic.

As a result of their adoption and security offering to large portions of the open web, CloudFlare’s integration of x402 is a major development for the structure of the internet. Websites that are increasingly inaccessible to the massive data demands from AI can now sell that data to AI agents for crypto. CloudFlare’s implementation only mentions Stablecoins such as USDC, the Open USD standard, but the protocol supports Bitcoin on-chain and is actively exploring integration of the Lightning network.

The Web is Broken

Kevin Leffew , co-author of the x402 protocol and AI GTM at Coinbase, told Bitcoin Magazine there’s a major user experience issue in the way AI currently interacts with the open web and x402 — which is now under the control of the Linux Foundation — is trying to solve it. “Every api call requires an api key, which in turn requires a human, and adds unnecessary friction,” Leffew explained, adding, “our goal is to kill the api key”.

Popular AI agents such as OpenClaw often require API access keys to special paid web search services, to let the AI agents access the web easily, with the mobility that a human user would enjoy. Services of this sort are offered by popular browsers and search engines such as Brave.com and Perplexity . But who out there wants to be paying a subscription service on top of computer hardware and internet access, plus AI token costs to search the web? These services also require a human to sign up with a credit card for a monthly subscription, paying for access that might be blocked by the websites holding valuable data anyway via non-standard methods. A better solution is needed.

AI agents need to be able to think about money and resource costs, and need to have a computer-friendly way to make payments for novel data. An example of this use case was recently demonstrated by an X account called “Lightning Mode AI,” which built a wrapper over ESPN FIFA data and had an AI agent pay for it in Bitcoin. The Agent was then able to quickly place bets on outcomes on markets like Polymarket, which could potentially let agent owners earn their money, during the soccer World Cup.

This example by Lightning Mode AI used an older implementation of the idea behind x402 called L402 , a protocol developed by Lightning Labs to specifically enable bitcoin payments for data on third-party websites.

Denial of service attacks (DOS) are also potentially solved by a machine native monetary system for the internet. The fundamental vulnerability exploited by these DOS attacks revolves around the bandwidth and computing costs to answer a question or query from an internet user. The user sends a request to view a website, the site’s server must compute, resolve and serve the website data back to the user; this has material costs at scale. DOS attacks send massive amounts of requests, often from malware-infected networks of machines (DDOS), targeting the server resources of their victims. This kind of attack can, in theory, be stopped by simply asking for payment from the user before spending the resources to respond to the user’s query. But the payments must be cost-effective and fast enough for the user experience demands of the digital age.

Protocols like x402 and L402 enable websites to paywall access to their valuable data, while teaching AI agents how to pay for access. No credit cards needed, no user data explicitly shared with payment networks, no ‘are you a bot’ captchas, no annoying account registrations you never use again, no subscription service for web search api key. Just pay for the data you consume.

The Economics of Micro Transactions

This micro-transaction market between machines is not a new idea. It has been theorized by luminaries of the cypherpunk age like Nick Szabo and others, though it has, up until now, been found lacking. Szabo argued that the biggest problem with micro transactions was not just payment technology, but the cognitive transaction costs involved.

Every time a user makes a payment, their brain needs to calculate whether it is worth it; this also has a cost on users, which can probably be measured in calories, and sometimes deciding to pay a couple of pennies for data is not worth the effort. But AI agents change this equation, in theory.

If AI becomes a new way for users to interact with the open way, then the cognitive costs involved in calculating the merit of spending pennies and even sub-penny values for data might be effectively abstracted away.

Users can simply give their bot a budget with spending policies and let it do its best to use that money wisely. Whether AI agents can be responsible enough to safeguard user funds remains to be seen, but some experiments demonstrate that AI agents can be reasonably resilient at the job. Take Freysa AI , for example, a 2024 era AI agent that held up against 48,000 prompt engineering attempts. Users paid to try to convince the bot to release funds in a smart contract treasury to them; if the bot refused, the bot kept the user’s money, adding it to the treasury. Eventually, someone managed to fool the bot , but not after $50,000 worth of attempts had been made. With hard-coded spending limits, the risk of prompt engineering an AI into giving way its web search lunch money is probably manageable.

The scalability of privacy-preserving digital payments in decentralized, censorship-resistant ways is also effectively already solved. According to Leffew, blockchains like Solana can do payments for a thousandth of a cent and settle it in milliseconds. Bitcoin’s Lightning network can also compete at the micro-transaction scale, and other Bitcoin protocols like the e-cash variants can be as fast as any internet packet, likely beating a highly centralized blockchain system like Solana.

Viktor Ihnatiuk, co-founder of UTEXO , told Bitcoin Magazine that they are actively working with the x402 developer community to integrate Bitcoin’s layer two protocols via RGB as a payment option. RGB integration would unlock layer two-style Bitcoin payments as well as USDT on Bitcoin settlements. The x402 standard, according to Leffew, is designed to be fundamentally neutral to the payment rails involved, even extensible to fiat rails, though likely will be dominated by cryptocurrencies and, for the foreseeable future, stablecoins.

Why CloudFlare Should Accept Bitcoin

CloudFlare’s x402 pilot program is a great step towards this vision of a cryptographic money actively used as the native currency of the internet. Its focus on stablecoins to start is also understandable, given the powerful brand and adoption of the dollar, which keeps accounting simple. However, there are a lot of underlying risks involved in how stablecoins are used today that Bitcoin solves.

For starters, most of the stablecoin volume moves on top of Ethereum Virtual Machine (EVM) style blockchains, which use an account model of public addresses; these are actively reused, creating long, detailed, public histories of financial engagement for each user. This is abysmal for user privacy, and tooling to obfuscate user flows on EVM blockchains, such as VPNs for crypto payments, are not common.

As a result, AI agents and their users are actively leaking data that might expose them to targeted attacks from organized cybercrime, among other risks.

Bitcoin, on the other hand, uses a UTXO model, where best practices lead people to create a new address for every payment received, resulting in payment trails that can be more difficult to track. Furthermore, Bitcoin’s fast payment protocols like Lightning, Ark or e-cash often deliver much greater privacy benefits to users by moving value off-chain via various smart contract-related technologies.

Last but not least, stablecoins are fundamentally anchored to the U.S. dollar and its foreign policy. If CloudFlare wants to be a viable option for the multipolar world, it will want to start taking a neutral stance on money. The dollar, while still the most valued currency in the world, is starting to lose ground to rising powers in the east , while alternative, geopolitically neutral currencies like Bitcoin continue to rise. Bitcoin might help CloudFlare maintain or even grow its position as critical internet infrastructure in the multi-polar world.

This post Cloudflare x402 Integration Opens Door for Bitcoin in AI Agent Micropayments first appeared on Bitcoin Magazine and is written by Juan Galt .