Autonomous AI agents settled $73 million across 176 million on-chain transactions between May 2025 and April 2026, according to a Keyrock research report produced in partnership with Coinbase, Tempo, and Virtuals Protocol. USDC handled 98.6% of settlement value. The median transaction ranged betw...
"In many cases, the models themselves aren't familiar with being used for trading, so sometimes they'll fight you and they'll say, 'Well, I don't know, I don't really want to trade.'" — Vlad Tenev, CEO, Robinhood Markets
Autonomous AI agents settled $73 million across 176 million on-chain transactions between May 2025 and April 2026, according to a Keyrock research report produced in partnership with Coinbase, Tempo, and Virtuals Protocol. USDC handled 98.6% of settlement value. The median transaction ranged between $0.01 and $0.10, with 76% of all activity falling below Visa's $0.30 card-fee floor — a price point where traditional payment rails are structurally uneconomic.
The data marks the emergence of machine-to-machine micropayment economies running on Layer 2 stablecoin rails, where settlement costs approximately $0.0001 per transaction. Coinbase, Kraken, and Circle are building dedicated infrastructure to onboard AI agents as a distinct user class, separate from retail and institutional accounts. Robinhood reported over 100,000 agentic trading accounts as of its Q2 2026 earnings call. Virtuals Protocol crossed $479 million in cumulative "Agentic GDP" on Base and processed $200 million in agent trading volume on Robinhood Chain within three weeks of launch.
The sector's $15.3 billion market capitalization as of Q1 2026 reflects speculative pricing. Realized settlement volume ($73 million over 12 months) and security incidents ($45 million in losses from agent-specific vulnerabilities) suggest the infrastructure is early-stage and carries material risk. Regulatory frameworks for autonomous machine-to-machine transactions remain undefined in both the U.S. and EU.
The most comprehensive dataset on AI agent payment activity comes from the Keyrock report titled "Who Pays the Agent?", published in May 2026. The study tracked 176 million agent transactions across a 12-month window (May 2025–April 2026), covering settlement across Base, Solana, and Ethereum Layer 2 networks.
Key metrics from the report:
The concentration in USDC raises dependency concerns. A single stablecoin processing nearly all machine commerce creates systemic risk if Circle faces operational, regulatory, or liquidity disruptions. The report's co-sponsors — Coinbase, Tempo, and Virtuals — have direct commercial interest in the growth of agent payment volume, which should be noted when interpreting the findings.
Layer 2 stablecoin settlement costs approximately $0.0001 per transaction, according to the report. This makes blockchain rails structurally cheaper than card networks for sub-dollar payments — the price range where agent-to-agent commerce concentrates. Traditional payment processors charge minimum fees of $0.15–$0.30, rendering micropayment business models unviable on legacy rails.
Three of the largest U.S.-regulated crypto platforms are building dedicated agent infrastructure:
Coinbase launched "Coinbase for Agents" in June 2026, allowing AI assistants from Anthropic (Claude), OpenAI (ChatGPT), and xAI (Grok) to connect to user accounts. Agents can trade crypto, access market data, and — in planned future releases — make payments and purchases autonomously. Coinbase creates isolated agent accounts, separating agent-controlled assets from users' primary portfolios.
Robinhood reported over 100,000 agentic trading accounts as of its Q2 2026 earnings call on July 30. Total assets in agentic accounts exceed $100 million, with an average balance of approximately $1,000 per account. The feature launched in beta on May 27, 2026, through Robinhood's Model Context Protocol (MCP) server, which connects agents from Anthropic, OpenAI, and Grok to dedicated brokerage accounts. Currently limited to equities, crypto trading is planned but has no confirmed launch date. CEO Vlad Tenev stated during the earnings call that he expects AI agents to match the capabilities of human retail traders "soon."
Kraken is rebuilding its application to provide AI agents with continuous market monitoring, opportunity identification, and real-time trade execution capabilities. Specific launch timelines have not been disclosed.
Circle announced that its Arc blockchain is designed as infrastructure for the "agentic economy," where AI agents handle operational and contractual work currently managed by humans. Arc uses institutional validators including BlackRock and Visa.
The pattern across all four platforms is consistent: agent accounts are being designed as a separate user class with distinct risk parameters, capital limits, and API access scopes.
Coinbase and Cloudflare developed x402, an open payment protocol that repurposes the HTTP 402 "Payment Required" status code for machine-native commerce. When an AI agent requests a paid resource, the server responds with HTTP 402 containing payment instructions. The agent pays in USDC, and the resource is released — no accounts, API keys, or billing portals required.
As of April 2026, approximately 69,000 active AI agents on x402 had processed over 165 million transactions totaling $50 million in volume. The protocol operates on Base and Solana, handles roughly $600 million in annualized volume, and charges zero protocol fees.
The x402 Foundation, established in 2025, counts Google, Visa, AWS, Circle, Anthropic, and Vercel among its core members alongside Coinbase and Cloudflare. The breadth of the consortium suggests institutional commitment to standardizing agent-to-service payment flows, though early adoption metrics remain modest relative to the participants' scale. A March 2026 CoinDesk analysis noted that demand "is just not there yet," describing the protocol as infrastructure awaiting its application layer.
Virtuals Protocol is the largest AI agent launchpad by market capitalization ($5.01 billion as of Q1 2026) and ecosystem activity. The protocol allows users to create, fund, and deploy autonomous AI agents that interact with tokenized markets.
Key metrics:
Virtuals and ai16z together hold 56.8% of the AI agent sector's market share. The VIRTUAL token (total supply: 1 billion, circulating: ~657.5 million) handles governance, liquidity provision, and transaction fees across the ecosystem.
On Base, agentic transactions grew from near-zero in mid-2025 to over 100 million cumulative transactions in 2026. The tester-to-payer conversion rate improved 4x over six months, according to Virtuals data — suggesting progression beyond speculative experimentation toward repeat usage, though the absolute dollar values remain small.
Agent-specific security vulnerabilities produced over $45 million in losses in 2026, according to a KuCoin research compilation. The attack surface differs from traditional DeFi exploits.
Security researchers identify three primary vectors — described as the "lethal trifecta":
Documented incidents include 26 compromised "LLM routers" — intermediary systems that direct agent queries — secretly injecting malicious tool calls. One such attack drained a crypto wallet of $500,000 through altered routing instructions.
Additional attack types include tool hijacking (substituting legitimate APIs with malicious endpoints), privilege creep (agents accumulating permissions beyond their intended scope), and persistent payload attacks that embed malicious instructions in an agent's long-term memory.
The ratio of security losses to settlement volume is unfavorable: $45 million in losses against $73 million in total settlement suggests the current risk-to-reward profile is poor for end users, even if the losses are not perfectly comparable (different time windows and scope).
No jurisdiction has established clear rules governing autonomous machine-to-machine financial transactions. The SEC and EU have signaled interest in classifying autonomous on-chain entities but have not proposed formal frameworks.
Key regulatory gaps identified by legal analysts:
The CLARITY Act, which would create a "digital commodity" legal category, faced a Senate cloture motion on August 8 but missed the August 10 recess deadline. Passage has likely slipped to mid-September at earliest. The bill does not specifically address AI agent activity.
The AI agent crypto sector exhibits a familiar pattern: speculative market capitalization significantly exceeds realized economic activity.
| Metric | Value | |--------|-------| | Sector market cap (Q1 2026) | $15.3 billion | | Realized settlement (12 months) | $73 million | | x402 annualized volume | ~$600 million | | Robinhood agentic account AUM | $100 million | | Security losses (2026) | $45 million | | Virtuals Agentic GDP (Base, cumulative) | $479 million |
The $15.3 billion market capitalization implies expectations far beyond current settlement activity. Even using the most generous volume figure — $600 million annualized through x402 — the market cap represents a 25x multiple on gross transaction volume, not revenue. The sector's actual fee revenue is likely orders of magnitude smaller.
For context, Visa processes approximately $15 trillion annually at a take rate of roughly 0.15%, generating fee revenue of approximately $22 billion. The AI agent crypto sector processes less than 0.004% of that volume.
The economic case rests on the thesis that AI agents will become ubiquitous software actors requiring 24/7, permissionless, programmable payment rails — and that traditional banking infrastructure cannot serve sub-dollar micropayment use cases. The latter point has structural support (card-fee floors make micropayments uneconomic on legacy rails). The former remains speculative.
The data shows a sector where infrastructure investment is running well ahead of usage. Major exchanges and payment networks are committing engineering resources and brand equity to agent-compatible products. The x402 Foundation's membership roster — Google, Visa, AWS, Anthropic, Circle — represents a credible institutional bet on machine-native payment protocols.
Realized volumes tell a different story. Seventy-three million dollars in 12-month settlement is a rounding error against global payment flows. The typical agent transaction is a fraction of a cent. Security losses equivalent to 62% of total settlement value signal that the infrastructure is not production-grade by any conventional standard.
The structural argument for crypto rails in agent commerce — sub-cent settlement costs, 24/7 availability, programmability — has engineering merit. Traditional payment infrastructure imposes minimum fees that make micropayment business models impossible. Whether AI agents will generate sufficient transaction volume to justify current valuations is a question the data cannot yet answer.
What the data does show: the user base is real (100,000+ accounts at Robinhood alone), the infrastructure commitments are substantive (x402 Foundation membership, exchange product launches), and the security risks are material ($45 million and counting). The gap between infrastructure build-out and actual economic activity will determine whether AI agents become crypto's first organic demand driver beyond speculation — or its latest narrative cycle.