Today on The Monday Signal: the SEC signals notice-and-comment rulemaking for on-chain markets, NEAR ships an AI agent marketplace with confidential GPU compute, and emerging-market data shows crypto exchanges are now functioning as banking infrastructure for 77% of users.
In a May 9 speech at the Special Competitive Studies Project's AI+ Expo, SEC Chair Paul Atkins committed to formal notice-and-comment rulemaking — rather than enforcement-by-interpretation — to clarify how exchange, broker-dealer, clearing-agency, and crypto-vault definitions apply to on-chain protocols. Atkins explicitly acknowledged that single protocols often combine multiple traditional intermediary functions and require fresh regulatory analysis rather than forced compliance with legacy categories.
Why it matters
This is the most concrete signal yet that the SEC is moving from prosecution-by-ambiguity to a function-based regulatory framework, with autonomous on-chain systems (including agent-driven finance) explicitly named as a target of clarification. For builders of decentralized AI agents and DeFi infrastructure, this matters in two ways: it opens a formal comment process where protocol-level arguments can shape definitions, and it pairs with the Clarity Act vote on May 14 to suggest the US is finally racing the EU and Singapore on enabling rules rather than restrictive ones. Watch for the actual NPRMs — speeches set direction, but the proposed text will determine whether 'function-based' becomes friendly or just differently restrictive.
Binance disclosed that emerging-market users now represent 77% of its platform (up from 49% in 2020), with stablecoins at the center of a usage shift away from trading toward savings, remittances, and payments. Adjacent data points published the same week reinforce the pattern: stablecoins drive 90% of Peru's $28B annual crypto volume with remittance costs collapsing from 6.6% to under 0.5%, stablecoin card spending is up 105% YoY led by LatAm, and Nigerian fintech Paga's Sui partnership extends from payments into tokenized bonds and real estate.
Why it matters
This is the structural shift that makes the agent-payment rails being built in parallel actually consequential — the user base for stablecoin infrastructure is overwhelmingly in markets where the alternative is no banking, not better banking. For someone running 64 chapters globally, the practical implication is that the highest-leverage community work is increasingly outside the US/EU bubble, and the local interface layer (rather than the underlying protocol) is where adoption is won. The Binance number and the Peru data together suggest the 'crypto as speculation' frame is now actively misleading for most of the actual user base.
On May 9, NEAR Foundation launched four coordinated products: a decentralized AI Agent Market for agent-to-agent transactions, NEAR AI Cloud (a confidential GPU marketplace running inside Trusted Execution Environments), the IronClaw assistant, and admission into NVIDIA's Inception Program. The marketplace supports Claude, OpenClaw, and Codex-class frameworks and uses NEAR's sub-second finality for settlement.
Why it matters
This is one of the few stacks shipping all three layers — agent discovery, confidential compute, and on-chain settlement — under one roof rather than as separate integrations. The TEE-based GPU layer is the specific piece that addresses the bottleneck most decentralized agent deployments have been hand-waving past: how to run frontier inference for paying customers without leaking either prompts or weights. Worth watching whether the NVIDIA credential translates into actual H100/B100 capacity flowing to the marketplace, or remains a logo on a slide. For DAIAA's mission frame, this is the closest thing to a reference architecture for agent proliferation on decentralized rails.
Circle released a developer guide and open-source reference implementation for Nanopayments on Arc, enabling agents to settle USDC payments down to $0.000001 at thousands of transactions per minute via off-chain authorization batched to on-chain settlement through Circle Gateway. The reference build includes a working agent that autonomously buys API calls.
Why it matters
This sits one layer below the AWS Bedrock AgentCore / x402 stack covered last week — where x402 handles discovery and request semantics, Nanopayments addresses the economics of metered M2M commerce that x402 alone can't make profitable. With Coinbase's x402, Google's AP2, and now Circle's Gateway-based Nanopayments all in production within a month, the agent payment stack is consolidating faster than most governance and identity layers. The interesting question is whether Arc becomes a settlement venue agents actually choose, or whether Nanopayments primitives get ported back to Base and Solana where existing x402 flows already live.
At Consensus Miami, Trust Wallet introduced an agent kit and confirmed it is implementing EIP-8004 for on-chain agent identity, while Mesh detailed Smart Funding — cross-chain payment routing designed to abstract chain selection for both human and agent users. The discussion explicitly addressed liability assignment, on-chain identity binding, and composability across wallet implementations.
Why it matters
EIP-8004 going from spec to a major retail wallet is the pragmatic on-ramp the agent identity stack has been waiting for. Most agent identity proposals so far have lived in research repos or testnets; Trust Wallet's distribution flips that. For builders trying to make agents legally addressable — KYA frameworks, audit trails, NVNM-style decision receipts — having a wallet-resident identity primitive matters more than another standards proposal. Watch whether other major wallets follow EIP-8004 or fragment into competing schemes.
China's Cyberspace Administration, NDRC, and MIIT jointly published implementation guidelines on May 8 for AI agent development under the 'AI plus' action plan. The framework defines agents as autonomous systems with perception, memory, decision-making, and execution; identifies 19 priority use cases across science, industry, and public services; and emphasizes 'safety and controllability' as co-equal with innovation. A parallel Substack interview with Xiaomi's Luo Fuli (ex-DeepSeek) describes Chinese frontier labs reorganizing compute allocation from roughly 3:5:1 toward 1:1:1 (research:pre-training:post-training) for the agent era.
Why it matters
Two adjacent signals say the same thing: the global frontier in AI competition is shifting from chat models to agent systems, and China is now formalizing the policy stack to push it. The guidelines' 'controllability' framing is the explicit philosophical opposite of the decentralized AI thesis — not regulation of agent harm, but state legibility of agent behavior. For DAIAA, the strategic implication is that decentralized agent infrastructure now has a clear positioning against both US enterprise lock-in and Chinese state-supervised deployment. The Luo Fuli interview is the more interesting tactical signal: if China's labs believe sophisticated frameworks let weaker open-source models compete with frontier closed ones, the decentralized + open-weight thesis just gained an unlikely validator.
Project Eleven followed Google Quantum AI's revised qubit estimate (~500,000 physical qubits to break 256-bit ECC, down ~20x from prior estimates) with a 110-page report formalizing a $2.3T at-risk figure across ~6.9M BTC in quantum-vulnerable addresses. The new detail: Taproot's default-exposed public keys are identified as a primary contributor to that exposure, putting a privacy/efficiency upgrade ratified in 2021 at the center of the threat model. Project Eleven's modeled attack window is ~9 minutes against a broadcast transaction key with a 41% pre-confirmation success rate.
Why it matters
The Google estimate was flagged in prior coverage; today's addition is the Taproot attribution and the 110-page migration-coordination argument. Project Eleven's core claim is that the binding constraint is governance, not cryptography — Bitcoin has never executed a non-opt-in migration affecting every wallet, exchange, and custodian. The Taproot finding is politically uncomfortable precisely because the upgrade was ratified with broad community support; it now becomes Exhibit A for the urgency camp's argument that PQ migration cannot wait for organic opt-in adoption curves.
Boltz launched non-custodial swaps between Bitcoin (including Lightning) and native USDC across Ethereum, Arbitrum, Base, Polygon and other chains, using Circle's CCTP for cross-chain delivery. No accounts, no KYC, no custody handoff.
Why it matters
The interesting piece isn't the swap — it's the on/off-ramp topology. Most non-custodial Bitcoin↔dollar flows today still terminate at a centralized exchange or licensed PSP that effectively re-introduces custody and surveillance. Boltz pairs Lightning-side privacy with Circle's regulated dollar leg via CCTP, which is the cleanest architectural answer yet to the 'regulated stablecoin without surveilled fiat ramp' problem. Useful primitive for Bitcoin-native payroll, merchant settlement, and any agent system that needs to bridge to USDC without giving up self-custody on the BTC side.
Manhattan federal judge Margaret Garnett modified the May 1 SDNY restraining notice on May 9 to allow Arbitrum DAO's Constitutional AIP vote to proceed — explicitly shielding vote participants from contempt liability — while keeping the underlying 30,765 ETH (~$71M) transfer barred. The carve-out distinguishes the governance act (protected) from execution (still blocked). Aave and other named executors remain personally exposed to contempt risk. This follows the Security Council's 9-of-12 emergency freeze on April 21 and the 90.96% Snapshot approval to release the funds.
Why it matters
The court has now drawn the precise line DAO governance designers have been dreading: voting is protected expressive activity, executing the vote is not. That's workable for Arbitrum but creates a durable operating hazard for any DAO treasury that could be touched by terrorism-judgment creditors, OFAC tracing, or third-party attachment. The named-executor contempt exposure — flagged in prior coverage — is now real, not theoretical. The design implication: constitutional processes need execution layers that can recognize and pause on external legal events without requiring another vote, rather than treating on-chain ratification as self-executing.
Ondo Finance integrated with Broadridge — which settles roughly $15 trillion daily in traditional shareholder communications — to enable holders of Ondo's 260+ tokenized assets ($920M TVL) to submit voting preferences, review filings, and receive shareholder communications directly through their wallets. It is the first at-scale deployment of on-chain shareholder voting for tokenized real-world equities.
Why it matters
Tokenized equities have been a five-year promise; the missing operational piece was almost always governance plumbing rather than the wrapper. Routing through Broadridge solves that with the incumbent that custodians and transfer agents already trust, which is also why this won't displace DAO-style governance — it extends shareholder rights onto chain rather than reinventing them. For watchers of governance innovation, the more interesting question is whether wallet-native voting UX changes participation rates relative to traditional proxy voting, which historically hovers near zero for retail.
Cardano delegate @ItsDave_ADA used 66.7M ADA in voting power to single-handedly reject a treasury withdrawal proposal seeking 13M ADA (~$3.1M) for upgrades, citing inadequate financial transparency and the lack of granular budget breakdowns. The veto is part of a broader pattern under Cardano's Intersect committees, which were seated last week using one-member-one-vote rather than token-weighted voting.
Why it matters
This is the kind of governance event that doesn't make the front page but should — a single delegate exercising fiduciary judgment to reject a bundled, opaquely-justified treasury ask. It's the inverse of the Gnosis whale flip: same concentration of voting power, opposite directionality, and a much healthier signal about how DAO budget review can actually function. For anyone running community treasuries, the concrete takeaway is that pre-vote budget granularity requirements are increasingly table-stakes — a reasonable answer to the 67% of DAOs still holding treasuries in native tokens with no real review process.
OpenAgents, a Bitcoin-native AI lab founded by Christopher David, closed a $1.3M pre-seed after graduating from the BitcoinFi accelerator. The company is scaling Pylon — a distributed compute node that pays contributors in BTC for spare compute — and Psionic, a Rust-based ML framework supporting inference, embeddings, fine-tuning, and distributed training on consumer hardware. The beta network has paid out over 1M sats across 1,000+ active Pylon instances.
Why it matters
This is exactly the kind of small round that's easy to miss but worth flagging — it's a working tokenized compute economy settling on Bitcoin rather than a token-of-the-week, and it targets consumer hardware rather than competing for hyperscaler GPU capacity. The interesting design choice is using Bitcoin/Lightning for settlement instead of a native token, which removes the speculative supply-side dynamics that have hollowed out most prior decentralized compute networks. Worth watching whether Psionic gets adoption from the open-weight model community as a serious local-first ML framework.
A technical synthesis maps the three production agent-payment protocols in active deployment: Google's AP2 (FIDO Alliance delegated authorization, identity-anchored), Coinbase's x402 (HTTP 402 + USDC, ~119M cumulative transactions, now embedded in AWS Bedrock AgentCore as of May 7), and Cloudflare/Stripe's autonomous cloud provisioning stack. Circle's Nanopayments reference implementation on Arc — enabling sub-$0.000001 USDC settlement at thousands of transactions per minute — adds a fourth layer below x402, handling the metered M2M economics x402 alone can't make profitable.
Why it matters
Useful framing for builders trying to figure out where to standardize: these protocols aren't really competing — they map to different trust models and target stacks. AP2 wins inside enterprise identity perimeters; x402 wins where agents are permissionless and chain-native; Cloudflare/Stripe wins where the agent and the workload share a cloud account. The interesting strategic question for decentralized AI is whether x402 can hold its 'neutral' positioning now that the Foundation lives at Linux Foundation but its primary integrator is AWS. The protocol fragmentation is real, but it's likely permanent — and that's actually fine, since most agents will speak more than one.
Connecticut passed SB5 (the Artificial Intelligence Responsibility and Transparency Act) on May 1 with bipartisan support, with Governor Lamont confirming he will sign despite federal pressure on states to defer to forthcoming federal frameworks. The law covers AI companions, synthetic-media transparency, automated employment tools, and frontier-model developer obligations — explicitly addressing autonomous AI systems and not just generative content.
Why it matters
Connecticut becomes the first state to legislate frontier-model and AI-companion accountability post the federal AI moratorium debate, and the explicit defiance of federal pressure matters for the broader agent regulation map. For decentralized AI agent deployments, the immediate risk is jurisdictional fragmentation — California's existing regime, the EU AI Act omnibus delay, China's new guidelines, and now Connecticut all impose subtly different definitions of what an 'autonomous system' is. KYA-style identity standards (ERC-8004, Visa TAP, etc.) will probably end up being the practical compliance layer that lets agents operate across these regimes, more than any single jurisdiction's rules.
Meghalaya Chief Minister Conrad K Sangma publicly rejected mass tourism in favor of phased, slow-tourism development — explicitly targeting longer stays, deeper community engagement, and environmental protection over visitor-volume growth. The 'samosa tourists' phrase refers to brief, low-spend visitors; the policy alternative aims for travelers who contribute meaningfully to local economies through extended residence.
Why it matters
A counter-narrative to the Instagram-era cultural-tourism critique landing the same week (Santiniketan) — Meghalaya is saying the quiet part out loud about ecologically sensitive destinations, and pairing it with policy rather than just commentary. The Vanuatu indigenous-women-led model covered last week sits in the same lineage. Worth tracking whether 'slow tourism' becomes a real regulatory category (visitor caps, minimum-stay incentives, community revenue-share mandates) or stays at the level of ministerial speeches.
Agent payment rails consolidate around three architectures Google's AP2 (FIDO-delegated), Coinbase's x402 (HTTP 402 + USDC), and Cloudflare/Stripe (cloud-native) are now the three protocols agent builders actually have to choose between. Circle's Nanopayments reference implementation on Arc and NEAR's AI Agent Market launch the same week effectively force the question for anyone building on decentralized infrastructure.
Emerging-market crypto adoption has crossed the 'banking infrastructure' threshold Binance reporting 77% of users from emerging markets, 90% of Peru's $28B volume in stablecoins, Paga/Sui partnership in Nigeria, and the Women in DeFi Summit's 6M-user target all point to the same thing: in much of the world, crypto rails are now the default banking stack rather than a speculative overlay.
DAO governance meets jurisdictional reality The Arbitrum/$71M ETH case — 90.96% on-chain approval, then a court order — is becoming the canonical precedent for what happens when DAO votes collide with off-chain legal authority. Combined with Cardano delegate fiscal vetoes and Mantle's cross-DAO credit facility, governance is shifting from theatrical voting to operational risk management.
China and the US converge on rulemaking, diverge on philosophy China's Cyberspace Administration issued formal AI agent guidelines emphasizing controllability the same week SEC Chair Atkins committed to notice-and-comment rulemaking for on-chain markets. Both jurisdictions are moving past enforcement-by-ambiguity, but the design philosophies — centralized oversight vs. protocol-function regulation — are diverging in ways that matter for where decentralized agents can actually deploy.
Post-exploit infrastructure consolidation favors Chainlink Solv ($700M), Kelp, Tydro, and re.al have all migrated to Chainlink CCIP within weeks. Aave's collateral-standards overhaul formalizes this as policy. The security gain is real; the centralization tradeoff — Chainlink now securing ~$32B and 58% market share — is becoming a governance question of its own.