The Agent Economy in CLARITY

July 10, 2026

The most visible structural changes in CLARITY (the trichotomy framework, the disclosure-only regime, the bank-permissibility expansion) are responses to problems the digital-asset industry has been pressing for years. The most forward-looking structural commitments in the bill are about a problem that is barely on the radar of most practitioners: the emerging intersection of autonomous AI agents and on-chain financial infrastructure. The agent economy is not a hypothetical. ERC-8004 (the agent-identity standard), x402 (the HTTP-payments protocol), EIP-7702 (the account abstraction provisions enabling smart-contract wallet behavior in externally-owned accounts), and the various agent-frameworks from Coinbase Agent Kit, Crossmint, Skyfire, and similar providers are all in active deployment. Autonomous trading bots have been a feature of cryptocurrency markets since 2014. The new class of LLM-powered agents that can hold wallets, sign transactions, participate in governance, and execute strategies against on-chain protocols is qualitatively different and has emerged faster than most regulatory frameworks have adapted.

Two Title V provisions of CLARITY are structurally important for the agent economy. §501 establishes a joint CFTC-SEC Micro-Innovation Sandbox enabling eligible firms to test innovative activities, with “innovative” explicitly defined to include artificial intelligence. §503 directs a GAO study of automated regulatory compliance, examining how distributed-ledger-based compliance tools and smart-contract functionality can satisfy regulatory obligations programmatically. Read together with the §301(a)(2)(C) activity carve-outs (which protect validators, sequencers, oracles, bandwidth providers, and similar infrastructure roles that AI agents may occupy), the bill establishes a structural framework that contemplates AI agents as legitimate participants in regulated financial markets.

§501: the Micro-Innovation Sandbox

§501(a)(4) defines “innovative” as “new or emerging technology, or a novel application of technology, including artificial intelligence, that (A) provides a financial product, service, business model, or delivery mechanism to the public; and (B) lacks (i) a substantially comparable, widely available analogue in common use in the United States; and (ii) an analogous Federal regulatory regime.” The explicit reference to AI is structurally important. The Sandbox is not just for blockchain-native innovation; it is for novel technology applications more broadly, with AI as a specifically enumerated category.

The Sandbox is established under §501(b) through joint notice-and-comment rulemaking by the SEC and CFTC, with a 360-day implementation deadline. The Sandbox operates “to enable eligible firms to test innovative activities within the United States, subject to (1) applicable Federal and State securities and commodities laws; (2) other State laws that are not specific to the regulation of securities or commodities; and (3) the limitations of this section.” The Sandbox is not a complete regulatory waiver; it operates against the existing regulatory framework with targeted exemptive relief for specific activities, subject to specific limitations.

§501(c) defines eligibility. A U.S.-based person is an eligible firm if it submits an approved application, seeks to conduct an eligible and lawful innovative activity, is not subject to specified statutory or state-law disqualifications, has no criminal conviction for fraud, agrees to SEC/CFTC jurisdiction, designates a point of contact, employs not more than 25 employees, and has annual gross revenues not exceeding $10 million. The size limitations target the framework at smaller firms and early-stage ventures, which is consistent with the “Micro-Innovation” naming and the broader policy goal of encouraging experimentation outside the existing regulatory channels.

§501(d) defines eligible activities and ceilings. The Commissions maintain and publish a list of eligible innovative activities, updated every two years through notice-and-comment, “sufficiently flexible to accommodate evolving technological developments, including distributed ledger-based products and services” and “focused exclusively on activities for which specific provisions of the securities and commodities laws may create a material impediment to the proposed innovative activity.” The activity list is the operational mechanism for defining what fits within the Sandbox; the structural choice is to leave it to the Commissions to identify the specific activities, with regular updates to keep pace with technological change.

§501(d)(2) establishes activity ceilings. An eligible firm cannot raise or commit more than $20 million in aggregate customer, investor, or counterparty funds in connection with Sandbox activities. The $20M ceiling, combined with the firm-size limits in §501(c)(1)(G)-(H), keeps the Sandbox at the experimentation-and-validation scale rather than the operating-business scale. §501(d)(3) imposes an annual participation cap of 20 projects per Commission.

§501(e) addresses the application process. An eligible firm submits an application describing the proposed innovative activity, identifying the provisions of the securities or commodity laws from which exemption is sought, explaining why relief is reasonably necessary, identifying material risks and mitigation, certifying compliance with anti-fraud laws, stating an exit objective involving Commission action, and agreeing to Commission jurisdiction and conditions. The Commissions have 180 business days to decide. The framework operates as a structured experimentation pathway with defined parameters and a defined exit.

§501(f) establishes a two-year participation period with a one-year extension where the firm is “actively pursuing the exit objective” and “making demonstrable progress.” The structural design is for time-limited experimentation followed by a defined exit (registration, exemptive order, interpretive guidance, no-action letter, or rulemaking petition). The Sandbox is not a permanent shelter; it is a temporary platform for developing the operational and regulatory framework that the participant will eventually operate under.

§501(g) addresses conditions and enforcement. The Commissions monitor Sandbox activities, coordinate supervision, and can disqualify a firm from continued participation for non-compliance. §501(k) provides limited preemption of state securities and commodities laws for Sandbox participants while preserving state enforcement of anti-fraud and general-applicability laws.

Why the Sandbox Matters for Agents

The Sandbox is structurally significant for the agent economy in three ways.

• First, the explicit AI inclusion in §501(a)(4) is the legislative signal that AI agents are within the framework’s intended scope. A firm deploying an AI-powered trading agent that operates against U.S. crypto markets, in a manner that would otherwise raise broker-dealer registration questions under §15(a) of the Exchange Act, can apply to the Sandbox for targeted relief. The firm’s two-year experimentation window is the period during which the operational reality of the agent’s activity can be developed and the regulatory pathway can be clarified.
• Second, the Sandbox addresses the absence-of-analogous-regime problem. AI agents operating on-chain do not have a clean regulatory analogue. They are not registered broker-dealers, not registered investment advisers, not banks, not money transmitters. The pre-CLARITY regulatory question was how to characterize these agents under the existing framework. The Sandbox provides a structural channel for working through that characterization in a controlled environment, with the Commissions’ active participation in defining the operational and regulatory parameters.
• Third, the Sandbox’s preemption of state law (§501(k)(1)) addresses the cross-state regulatory friction that would otherwise impede agent deployment. State securities and commodities laws, money-transmitter laws, and similar state-law frameworks would otherwise apply to agent activity on a state-by-state basis. The Sandbox provides limited federal preemption for the duration of participation, with state anti-fraud authority preserved. The structural choice is to consolidate the regulatory dialogue at the federal level for innovative activity.

The practical limitations are real. The $20M activity ceiling, the 25-employee and $10M revenue firm-size limits, and the 20-projects-per-year participation cap mean the Sandbox is not a vehicle for large-scale commercial deployment. It is a vehicle for early-stage experimentation. Firms developing agent products that intend to reach commercial scale will need to use the Sandbox to develop the regulatory framework and then exit to a permanent operating status (registration, exemptive order, or rule change). The framework operates as a regulatory escalator rather than as an indefinite shelter.

§503: Automated Regulatory Compliance

§503 directs the GAO to study automated regulatory compliance, defined as “the use of technology, including data standards, automation, and distributed ledger or smart contract functionality, to automate, tag, or otherwise streamline regulatory reporting, disclosure, supervisory, or other compliance obligations.” The study examines distributed-ledger-based compliance tools, the feasibility of satisfying regulatory obligations through on-chain code-based mechanisms, interoperability across agencies, data-collection systems, and standards or taxonomies that could support interoperability.

The substantive direction of §503 is significant. The bill contemplates that regulatory compliance can be implemented programmatically, with smart contracts and distributed-ledger records serving as the operational mechanism. The structural concept is that an issuer’s §4B(d) disclosure obligations, an intermediary’s recordkeeping obligations under Rule 17a-3 and 17a-4, an AML obligor’s SAR-filing obligations under §5318, and similar compliance obligations could be satisfied through on-chain mechanisms rather than through traditional reporting infrastructure.

The GAO study is the precursor to substantive regulatory implementation. The structural sequence is: study → pilot programs (recommended in §503(b)(2)) → guidance, rule changes, or statutory amendments. On a mid-2026 enactment, the 1-year study deadline produces a public report by mid-2027 that will inform the Commissions’ and other agencies’ subsequent rulemaking. Expect the §503 framework to generate substantive regulatory changes in the 2027-2030 timeframe.

The agent-economy relevance is direct. AI agents operating in regulated markets generate compliance obligations: order-reporting, trade-reporting, position-tracking, AML monitoring, and similar functions. Implementing those obligations programmatically through smart contracts and on-chain records is operationally simpler than retrofitting agent activity into traditional compliance infrastructure. The §503 framework is the legislative pathway for that programmatic implementation.

§301(a)(2)(C) and Agent Infrastructure Roles

Post #6 of this series walked through the §301(a)(2)(C) activity carve-outs: compiling network transactions, relaying, searching, sequencing, validating; providing computational work, operating a node or oracle service, providing bandwidth; participating in incident-response or security councils. The carve-outs are structurally important for the agent economy because AI agents may occupy any of these infrastructure roles.

Validators running AI-powered slashing-avoidance and reward-optimization software are within the §301(a)(2)(C)(i) carve-out. MEV searchers running ML-based mempool scanning and arbitrage strategies are within the carve-out. Oracle operators running ML-based data-quality systems are within §301(a)(2)(C)(ii). Block builders running AI-powered bundle-optimization are within (i). The structural commitment is that AI-powered infrastructure roles do not flip a protocol or its components into non-DeFi status simply by virtue of being AI-powered.

The combined effect of §301(a)(2)(C) and §501 is a framework that contemplates AI agents as both infrastructure providers (within the activity carve-outs) and as market participants (within the Sandbox). The structural choice is permissive: agents can run the infrastructure, agents can participate in the markets, and the regulatory framework adjusts to accommodate the activity rather than treating the agent’s automation as a categorical concern.

ERC-8004 and x402 in the CLARITY Framework

ERC-8004 is the Ethereum standard for agent identity, defining how AI agents can have on-chain identifiers, reputational records, and access controls. x402 is the HTTP-payments protocol that allows agents to make payments in stablecoins or other digital assets through web-standard interfaces. Both are operationally significant for the emerging agent economy and both raise structural questions about how the underlying activity is regulated.

An agent with an ERC-8004 identity operating against U.S. crypto markets, accepting payments through x402, and executing trades on U.S. exchanges raises a series of regulatory questions under current law. Is the agent a broker-dealer? An investment adviser? A money transmitter? Does the agent’s principal have liability for its actions? The current answers are uncertain.

The CLARITY framework would provide structural answers, mostly through the activity-based application of existing regulatory categories. If the agent’s activity is securities-related (executing trades on registered exchanges, making investment recommendations), the relevant securities laws apply on the same terms as they would to a human acting in the same capacity. The §301 activity-based application “without regard to technological form” reaches the agent on the basis of what it does, not on the basis of being an AI.

The structural permission for the activity is implicit in the bill’s framework. AI agents operating non-securities-related infrastructure roles (validation, oracle services, sequencing) are within the §301(a)(2)(C) carve-outs. AI agents operating securities-related activities (brokerage, advisory) are subject to the existing securities-law framework. AI agents operating novel financial activities without a clean analogue can use the §501 Sandbox to develop the regulatory pathway. The framework is comprehensive, even if the specifics will emerge through rulemaking and practice.

EIP-7702 and Account Abstraction

EIP-7702 enables externally-owned accounts (the standard user wallet type on Ethereum) to delegate execution to smart contracts, effectively giving EOA wallets some of the programmability of smart-contract wallets without requiring users to migrate to entirely new wallet architectures. The EIP is operationally important for the agent economy because it enables more sophisticated programmable behavior at the wallet layer.

An agent operating through a 7702-enabled EOA can execute programmatic transactions, batch operations, and complex multi-step strategies in ways that were previously available only through smart-contract wallets (ERC-4337 architecture). The structural significance for CLARITY is that the wallet operator (the user) remains in control of the wallet and the agent operates as a programmable layer within the user’s control framework, not as a separate custodial relationship.

This matters for §307 (self-hosted wallets, covered in Post #14) and §605 (Keep Your Coins Act). A 7702-enabled wallet is self-hosted under the §307(a)(1) and §605(b)(2) definitions: the user retains independent control. The agent’s programmatic activity operates within that control framework. The structural conclusion is that 7702-enabled agent-driven wallet activity is within the self-hosted wallet protections of CLARITY, even when the agent is executing complex programmatic strategies.

International Sandbox Comparisons

§502 of CLARITY provides for international cooperation, including the possibility of cross-border regulatory sandboxes building on the §501 framework. The international sandbox landscape has matured substantially. The UK Financial Conduct Authority’s Innovation Sandbox, launched in 2016, has supported hundreds of firms through structured experimentation pathways. The Monetary Authority of Singapore’s FinTech Regulatory Sandbox has supported similar experimentation in the Singapore market. The Hong Kong Securities and Futures Commission has operated comparable frameworks. The structural comparison is that CLARITY’s Sandbox is calibrated against these international precedents and operates in a similar structural mode.

The §502 cross-border-sandbox provision is operationally significant. A firm developing an agent-based financial product that intends to operate in multiple jurisdictions can pursue Sandbox participation in the United States under §501, while pursuing parallel participation in the UK FCA Sandbox, the Singapore MAS Sandbox, or other international frameworks. The §502 cross-border provision enables information-sharing and coordinated supervision across these frameworks, reducing the friction of multi-jurisdictional experimentation.

The structural lesson is that CLARITY would position the United States as a meaningful participant in the international experimentation landscape for innovative financial technology, including agent-based products. The pre-CLARITY posture, in which the U.S. regulatory framework was hostile to digital-asset innovation and the international leadership was in the UK and Singapore, would be structurally changed. The §501 Sandbox provides a domestic framework that operates alongside the international frameworks.

Positioning for the Agent Economy

Four planning consequences follow for firms building in this space.

• First, the §501 Sandbox is the structural pathway for early-stage experimentation that would otherwise face regulatory uncertainty. The framework is calibrated for early-stage firms ($10M revenue, 25 employees, $20M activity ceiling). Firms approaching commercial scale should plan to exit the Sandbox through one of the §501(e)(1)(F) exit pathways (registration, exemptive order, interpretive guidance, no-action letter, or rulemaking petition). The Sandbox is not an indefinite operating posture.
• Second, the §301(a)(2)(C) activity carve-outs cover AI-powered infrastructure roles. Agents running validation, oracle services, sequencing, MEV searching, or similar infrastructure activities are within the safe harbor regardless of whether the activity is AI-powered. The structural commitment is to activity, not to technology.
• Third, the §503 automated-compliance framework is the long-term pathway for programmatic implementation of regulatory obligations. The GAO study, due within a year of enactment, will inform subsequent regulatory changes. Firms developing on-chain compliance tools should engage with the GAO study process and prepare to engage with the subsequent rulemaking cycle.
• Fourth, the international landscape matters. The §502 cross-border-sandbox framework provides operational support for multi-jurisdictional experimentation. Firms with international ambitions should consider parallel participation in U.S. and international sandboxes.

The agent economy is at an inflection point. The infrastructure (ERC-8004, x402, EIP-7702, account abstraction generally) is maturing. The capital is available. The market opportunity is substantial. The remaining variable has been regulatory certainty, and CLARITY’s §501 and §503 frameworks would supply the structural commitments that enable the next round of experimentation. The bill is forward-looking on this topic in a way that most market-structure legislation is not, and the Title V provisions deserve to be read as substantive regulatory infrastructure rather than as throwaway sense-of-Congress provisions.

Written by David Lopez Kurtz