Chocolate & peanut butter, or nuts & gum?
Artificial intelligence (AI) agents will get a lot of use out of stablecoins, but a new study says ‘crypto’ boosters are dreaming if they think their technology will make these AI agents truly autonomous.
On June 8, The Initiative for Cryptocurrencies and Contracts issued a new 155-page survey titled Crypto x AI, AI x Crypto that pondered “what AI can do for blockchain-based technologies … and vice versa.”
The survey was released at an opportune moment, arriving just days after the privacy-focused Zcash token (ZEC) lost roughly half its value over a 24-hour period. The source of the crash? A ‘white hat’ security researcher used Anthropic’s Claude Opus 4.8 to detect a critical vulnerability in Zcash’s code that would have allowed less scrupulous researchers to mint an unlimited supply of ZEC.
While Zcash publicly disclosed the flaw only after it had been patched, many ZEC holders were sufficiently rattled to dump their holdings. And while ZEC’s value has since reclaimed much of its lost ground—thanks in part to Monday’s announcement that a new security audit of the Zcash protocol by Anthropic’s Mythos model had revealed no other potentially fatal flaws—it’s still sitting ~13% below the price it enjoyed prior to the announcement.
The revelation has other protocols and token-holders nervously awaiting news of other discoveries by ‘black hat’ researchers armed with the latest AI tools. Given the zeal with which state-sponsored hackers like North Korea’s Lazarus Group have targeted blockchain platforms, it may be that AI-fueled compromises are something we just have to get used to.
As the crypto/AI survey notes, going forward, AI will be deployed in the original design of new protocols/projects to minimize anomalies of this type. Of course, AI’s infamous capacity for hallucinations means human eyes won’t be rendered redundant anytime soon.
But the survey authors nonetheless expect “it will become the norm for smart contracts to be written largely by AI, possibly with the assistance of agentic frameworks.” And developers will “increasingly design not only blockchain algorithms, but also full-blown implementations with AI.”
In terms of what benefits crypto might offer AI, the authors seek to bust several ‘misconceptions’ that have been embraced by blockchain advocates. For instance, the authors cast doubt on the claim that blockchains can help distinguish human-generated content from material produced by generative AI tools, because “while a blockchain can guarantee the integrity of a record … it cannot guarantee that the information was true at the moment it was recorded.”
Blockchains are also deemed unsuitable for solving AI’s ‘bias and fairness problems.’ Blockchains “can indeed encourage transparency at various stages of the AI life cycle and broaden participation in AI governance,” but “the impact of these properties on downstream model outcomes remains unclear.”
Giving AI agents access to digital wallets is one of the more harmonious marriages of these two technologies and offers a level of automation to economic interactions, without the need for human input. But “automation should not be confused with autonomy: merely possessing a wallet does not make AI agents independent of human control.”
The authors note that “automated payments do not require blockchains; similar functionality can exist in centralized financial systems.” However, “blockchain-based payment systems can offer appealing properties such as neutrality and censorship resistance, which can be desirable for applications where payment suppression, censorship, or other forms of manipulation are a concern.”
Ripple tries to boost RLUSD with AI starter kit
While automated payments might not need blockchains, agentic AI payments involving blockchain-based stablecoins are becoming all the rage among both crypto and tradfi firms, with neither side looking to get left behind in this race.
Ripple Labs, issuer of both the XRP token and the RLUSD stablecoin, launched its XRPL AI Starter Kit last week, pitching it as “a new set of tools and integrations designed to help developers build agentic payment applications on the XRP Ledger (XRPL).”
The kit is launching in phases, starting with tools to help developers get familiar with building agent-powered applications. Once up and running, XRPL agents can engage in x402-powered payments in either XRP or RLUSD for API calls, AI model inference, and more.
Ripple launched RLUSD 18 months ago, but the token has struggled to gain traction in a field dominated by market-leader USDT (Tether) and Circle’s (NASDAQ: CRCL) USDC. RLUSD’s $1.65 billion market cap is ~$400 million higher than at the start of the year, but that remains a tiny fraction of the big boys’ caps. Still, hope springs eternal, as RLUSD debuted on the Gate.io exchange on Monday as one-half of a trading pair for USDT, BTC, ETH, and (surprise) XRP.
Coinbase for Agents, Base MCP
The x402 standard began life as a Coinbase (NASDAQ: COIN) project, which could explain why agentic AI payments have to date largely been conducted using USDC, from which Coinbase derives a significant share of its revenue. So don’t be shocked to learn that Coinbase continues to push this envelope.
Last week, Coinbase launched Coinbase for Agents (CFA), a link between your favorite large language model (LLM) and your Coinbase account. Users can allow the LLM to make trades on their behalf (within user-defined limits) while Coinbase is promising that CFA will “soon” allow x402-based payments.
The launch follows last month’s announcement of a near-identical set-up on Coinbase’s Ethereum layer-2 network Base. Base MCP (Model Context Protocol) similarly interfaces with popular LLMs to engage in all manner of transactions on its network, while also letting agents interact with Base apps and protocols (Uniswap, Morpho, etc.) via ‘skill plugins.’
Coinbase’s head of AI product, Lincoln Murr, told Fortune that Base users can use MCP to empower agents to conduct a wider variety of transactions across a number of different platforms. And when the agent wants to make a transaction according to your directions, it sends you a link to either confirm or cancel before any value is transferred.
That ‘confirmation required’ feature is a definite plus in a world in which the agentic kinks are still being worked out and the bad habit of “blind goal-directedness” has yet to be beaten out of our new digital slaves.
Mastercard moving at machine speed
Robinhood (NASDAQ: HOOD) began allowing its customers to connect their own AI agents to its trading platform last month, with the option to give your agent its own limited account so that it doesn’t go rogue and max out your credit card buying porn star memecoins.
On June 8, the popular Ethereum-based MetaMask digital wallet opened an early access program to its MetaMask Agent Wallet for select traders and developers. The new self-custodial wallet allows users to authorize agents to access swaps, perpetual futures, prediction markets, and a whole lot more across Ethereum Virtual Machine (EVM) networks and the HyperLiquid decentralized exchange (DEX). Here, too, users can set boundaries so their agents don’t go rogue.
Mastercard (NASDAQ: MA) has been among tradfi’s more enthusiastic adopters of stablecoin payment options, including spending $1.8 billion in March to acquire stablecoin infrastructure firm BVNK. The company recently launched a new Crypto Partner Program featuring “more than 85 crypto‑native companies, payments providers, and financial institutions” to collaborate on this brave new world.
Last week, Mastercard announced the launch of Agent Pay for Machines (AP4M), an extension of the Agent Pay program that the credit card giant launched last year. AP4M aims to empower AI agents to interact with each other “continuously at high velocity, executing chains of transactions, including microtransactions” involving “multiple payment types, from cards to stablecoins.”
Mastercard says it wants to empower businesses to “create services for AI agents to buy and use,” allowing transactions that are “programmatic, always-on and executed between systems in the background of digital commerce.” In a phrase that isn’t at all unnerving, Mastercard says the world requires “secure, reliable payments as software begins to transact on its own.”
To give AP4M a fighting chance, Mastercard has enlisted the support of a number of prominent crypto firms, including BVNK, Coinbase, Anchorage Digital, the OKX exchange, payment processor Stripe, and its Tempo layer-1 stablecoin payments network (to name just a few), as well as a handful of more traditional financial operators.
You must be this scalable to ride these rails
As Mastercard’s announcement noted, the fact that agentic AI payments will rapidly increase in both speed and volume “creates a new requirement: infrastructure that can keep up.” Stripe said as much earlier this year, warning that the demands of agentic AI will require “blockchains that support more than one million—or even one billion—transactions per second.”
Many of the networks promoting themselves as the preferred channels for agentic AI transactions may have initial success as a proof of concept. But skeptics believe these platforms will not only bend but flat-out break if they’re ever tasked with the tsunami of expected volume as individuals and entities around the world embrace the agentic AI revolution.
The real agentic AI crown will be worn by a truly scalable network that can expand to meet whatever demands are made of it, and do so while maintaining transaction fees at the same micro level of the payments being made.
In order for artificial intelligence (AI) to work right within the law and thrive in the face of growing challenges, it needs to integrate an enterprise blockchain system that ensures data input quality and ownership—allowing it to keep data safe while also guaranteeing the immutability of data. Check out CoinGeek’s coverage on this emerging tech to learn more why Enterprise blockchain will be the backbone of AI.
Watch: Can we trust AI? How blockchain and IPv6 could fix accountability
title=”YouTube video player” frameborder=”0″ allow=”accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share” referrerpolicy=”strict-origin-when-cross-origin” allowfullscreen>
