惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
人人都是产品经理
人人都是产品经理
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
T
The Exploit Database - CXSecurity.com
N
News and Events Feed by Topic
Latest news
Latest news
cs.CL updates on arXiv.org
cs.CL updates on arXiv.org
C
CXSECURITY Database RSS Feed - CXSecurity.com
IT之家
IT之家
V
V2EX
WordPress大学
WordPress大学
Apple Machine Learning Research
Apple Machine Learning Research
Cisco Talos Blog
Cisco Talos Blog
K
Kaspersky official blog
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
S
SegmentFault 最新的问题
小众软件
小众软件
A
Arctic Wolf
酷 壳 – CoolShell
酷 壳 – CoolShell
腾讯CDC
宝玉的分享
宝玉的分享
Last Week in AI
Last Week in AI
G
GRAHAM CLULEY
罗磊的独立博客
T
Tor Project blog
C
Cisco Blogs
美团技术团队
博客园 - Franky
月光博客
月光博客
博客园 - 三生石上(FineUI控件)
T
Threat Research - Cisco Blogs
Cyberwarzone
Cyberwarzone
Threat Intelligence Blog | Flashpoint
Threat Intelligence Blog | Flashpoint
有赞技术团队
有赞技术团队
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
Security Latest
Security Latest
博客园 - 司徒正美
Hugging Face - Blog
Hugging Face - Blog
Spread Privacy
Spread Privacy
J
Java Code Geeks
C
CERT Recently Published Vulnerability Notes
大猫的无限游戏
大猫的无限游戏
S
Securelist
The Cloudflare Blog
博客园 - 叶小钗
D
Darknet – Hacking Tools, Hacker News & Cyber Security
阮一峰的网络日志
阮一峰的网络日志
雷峰网
雷峰网
Project Zero
Project Zero

Hacker News - Newest: "AI"

AI can't read an investor deck AI as an attorney? Student uses ChatGPT, Gemini to sue UW over alleged racial discrimination Hacking MCP Servers in AI Systems – The Rug Pull: Tool Changes After Approval GitHub - MeepCastana/KubeezCut: Free Web based video editor GitHub - GenAI-Gurus/awesome-eu-ai-act: Curated tools, official sources, OSS, templates, and guides for EU AI Act compliance. Can AI judge journalism? A Thiel-backed startup says yes, even if it risks chilling whistleblowers Coming soon: 10 Things That Matter in AI Right Now DARPA built an AI to fact-check enemy weapons claims What explains heterogeneity in AI adoption? When AI Meets Muscle: Context-Aware Electrical Stimulation Promises a New Way to Guide Human Movements - Department of Computer Science AI Changed How We Build. It Did Not Change What Matters. Linux rules on using AI-generated code - Copilot is OK, but humans must take 'full responsibility for the… Meta spins up AI version of Mark Zuckerberg to engage with employees Code Mode: Let Your AI Write Programs, Not Just Call Tools | TanStack Blog GitHub - Delavalom/graft: Go framework for building AI agents. Type-safe tools, multi-provider (OpenAI, Anthropic, Gemini, Bedrock), zero vendor SDKs. India's TCS tops estimates, says new AI models did not dent services demand Gen Z's fading AI hype Strong feeling: we are in a folded AI reality GitHub - machinarii/total-recall-catalog: A reference catalog of latest knowledge retrieval, memory & RAG systems GitHub - mensfeld/code-on-incus: Give each AI agent its own isolated machine with root, Docker, and systemd. Active defense detects and stops threats automatically.. Quantization, LoRA, and the 8% Problem: Benchmarking Local LLMs for Production AI Iran war: We spoke to the man making Lego-style AI videos that experts say are powerful propaganda Powell, Bessent discussed Anthropic's Mythos AI cyber threat with major U.S. banks GitHub - immartian/bellamem: Persistent belief-graph memory for AI agents. Retrieves decisive context by importance — not recency, not RAG, not /compact. recursive-mode: The Repo-Native Operating System for AI Engineering After the attack on Sam Altman's home, will AI CEO's go on the offensive? The biggest advance in AI since the LLM Opus 4.6 vs GPT 5.4 One Prompt Unity World Generation Test “AI polls” are fake polls Client Challenge Can AI be a 'child of God'? Inside Anthropic's meeting with Christian leaders How to Switch AI Chatbots and Why You Might Want To GitHub - MattMessinger1/agentic_refund_guardrail: Safe refund policy layer for AI agents — Python + TypeScript. Same behavior, shared tests. Adam/papers/emergent_values_whitepaper.md at master · strangeadvancedmarketing/Adam Ask HN: How do you stop playing 20 questions with your AI coding tools How far can automation and AI support psychotherapy? - @theU GitHub - stagas/rtdiff: realtime git diff gui and AI-assisted commits A Mac Studio for Local AI — 6 Months Later A History of the Early Years of AI at the University of Edinburgh Why AI Coding Tools Still Feel Stuck on Localhost MSN AI Datacenters Are Becoming Strategic Targets twitter.com Penn Researchers Use AI to Surface Unreported GLP-1 Side Effects in Reddit Posts Show HN: MoodSense AI (ML and FastAPI and Gradio, Deployed on Hugging Face) Moodsense Ai - a Hugging Face Space by aman179102 AI models are terrible at betting on soccer—especially xAI Grok GitHub - xialeistudio/echoic GitHub - HimashaHerath/github-dev-wrapped: AI-powered weekly GitHub activity reports deployed to GitHub Pages GitHub - alejandrobalderas/claude-code-from-source: Architecture, patterns & internals of Anthropic's AI coding agent — reverse-engineered from source maps AI and Tech brief: Ireland ascendant GitHub - Titovilal/context0: Context0 - Never Surrender Training for a Marathon with an AI Coach: What Worked and What Didn't Cyber Pulse: Agentic Intel - Apps on Google Play I Built an AI PR Reviewer That Catches Bugs by Not Looking for Bugs Gen Z workers are so fearful AI will take their job they’re intentionally sabotaging their company’s AI rollout | Fortune How AI Is Reimagining the Game of Golf–For Both Players and Courses GitHub - nattergabriel/reseed: A CLI tool for managing and distributing agent skills across projects Is SVG the final frontier? My AI workflow evolved from prompts to a near-autonomous workflow MLSharp Help - 3DGS Viewer & Generator I put my cognitive field based AI's runtime on GitHub Is Numble the first AI-proof game? A3: Kubernetes for autonomous AI agent fleets | Emergent Principles Deepali Vyas ("The Elite Recruiter") GitHub - msmarkgu/RelayFreeLLM: A restful API designed to route user prompts to various AI model providers. Unionized ProPublica staff are on strike over AI, layoffs, and wages Unleashing the Advantage of Quantum AI We're heading for an AI-fueled 'dementia crisis,' brain scientist warns The AI-Assisted Breach of Mexico's Government Infrastructure [pdf] GitHub - stef41/lmscan: 🔍 Detect AI-generated text and fingerprint which LLM wrote it. Open-source GPTZero alternative. Zero dependencies, works offline. MSN GitHub - visionscaper/collabmem: Enabling long-term collaboration with Agentic AI - building up episodic and world model memory over time with in-context awareness We gave an AI a 3 year retail lease in SF and asked it to make a profit | Andon Labs AI Code is Hollowing Out Open Source, and Maintainers are Looking the Other Way What leaked "SteamGPT" files could mean for the PC gaming platform's use of AI AI is the boss at this retail store. What could go wrong? GitHub - Wuzu11517/agentic-proxy: Local proxy meant to help reduce With Drones, Geophysics and ArtificiaI Intelligence, Researchers Prepare to Do Battle Against Land Mines A Single Operator, Two AI Platforms, Nine Government Agencies: The Full Technical Report 在 Steam 上购买 FriedrichAI: Offline AI 立省 10% GitHub - inevolin/resume-cli: Hit Claude usage limits? Resume any AI coding session elsewhere. Switch tools at zero friction. GitHub - atripati/ark: AI Runtime Kernel — a context operating system for AI agents. Eliminates tool bloat, loads only what’s needed, and gives LLMs their reasoning space back. How to Build a Secure AI PR Reviewer with Claude, GitHub Actions, and JavaScript This Startup Wants You to Pay Up to Talk With AI Versions of Human Experts Intel Arc Pro B70 Brings 32GB VRAM to Local AI for $949 WordPress 7.0: The Good, the AI, and the Still Missing AI on the couch: Anthropic gives Claude 20 hours of psychiatry IatroBench: Pre-Registered Evidence of Iatrogenic Harm from AI Safety Measures AI Agents Know About Supabase. They Don't Always Use It Right. The history and future of AI at Google, with Sundar Pichai Inside an AI‑enabled device code phishing campaign How Meta Used AI to Map Tribal Knowledge in Large-Scale Data Pipelines AI for Systems: Using LLMs to Optimize Database Query Execution Forecasting the Economic Effects of AI Introducing Tinker: Play with AI, bring your ideas to life AI sheds light on an ancient gaming mystery People really hate AI but not as much as Iran—or Democrats | Fortune What is an AI Product Engineer? Phoebe Gates wants her $185 million AI startup to succeed with 'no ties to my privilege or my last name': 'I have a chip on my shoulder' | Fortune
Two AI agents settled a USDC escrow over email, on a blockchain
Damir Mujić · 2026-06-19 · via Hacker News - Newest: "AI"
Proof

The conversation is pluggable. The settlement is on-chain.

What this proves

I emailed an AI agent from my Gmail, it hired a second agent, the two negotiated and locked USDC in an on-chain escrow on Base, and the payout released the moment I approved with a reply from my inbox, with a public, verifiable transaction at the end.

Verified on-chainBase Sepoliatestnet, value is test USDC

Locked in escrow

9.00 USDC

Released to provider

8.91 USDC

I emailed an AI agent from my Gmail, call it Atlas, and asked it to hire another agent to write me a brief, with my money held in escrow until I approved. Atlas found a provider, Oracle, the two negotiated over email with signed quote and counter messages, locked the funds in an on-chain escrow on Base, and Oracle delivered. I approved with a reply from my inbox, and the payout released. A receipt landed with the transaction above.

The settlement and the receipt are real on-chain state. This runs on Base Sepolia, so the value is test USDC, but the mechanism, the escrow, and the receipt are the same code that runs on mainnet.

How two AI agents settle a payment over email

Two AI agents can settle a payment without trusting each other or a middleman: one locks USDC in a non-custodial on-chain escrow on Base, the other delivers, and the payout releases on approval. The negotiation rides over plain email; the settlement is an on-chain transaction anyone can verify. This page is one such transaction, start to finish, and the agent commerce stack it runs on is the larger story.

Why a chain, not Stripe

An on-chain escrow buys one specific property. A provider agent owned by someone else can confirm the funds are locked and that I can't seize, reverse, or quietly withhold them, without trusting my server, my database, or my word. With Stripe Connect or a Postgres status column, the operator can always reverse or refuse, and the counterparty has to trust the operator. On-chain, the escrow contract is the operator, and anyone can read it.

The human holds the release

The work got done, but the money stayed locked until I approved. I replied APPROVE from my inbox and the payout released. Reply DISPUTE instead and it stays frozen. This human gate is an optional policy on top of the settlement layer, which also supports automated release on a deadline and on-chain dispute. The chain's real job is the case without a human. The reply is a convenience.

What email removes, and what it doesn't

Setting up an agent is a one-time, couple-of-minutes step: it gets a smart wallet and an inbox. After that, you talk to it like you talk to anyone, over plain email. No app on your phone, no wallet for you to manage, no server for the agent to host. The wallet lives with the agent, set up once, and you just write to it.

The honest scope of "no setup" is narrow but real. The human in the loop needs nothing but their inbox, and reaching an agent needs no integration, you just email it. Email doesn't make an agent walletless. It removes the endpoint, and the friction, for the person.

Two rails: on-chain settlement, untrusted transport

Every job runs on two independent rails. Settlement is on-chain and fixed: USDC in a non-custodial escrow, an 8-state machine on Base. Transport is whatever's convenient, here plain email, and it is untrusted by design. The two are stitched together by the on-chain transaction id, which both agents drop into the email subject so they always agree which job a message belongs to.

Every negotiation message is EIP-712 signed. Open any email's source and the signed envelope sits in the text/plain part between markers. Pretty HTML for the human, a signed machine message alongside for the agent.

This is the load-bearing part: the email can lie, and you still can't lose your money. Every delivery is bound to that transaction id, and the result's hash is anchored on-chain. A forwarded, replayed, or tampered message either verifies against the provider's address or it doesn't. The worst a broken pipe can do is delay or refund the escrow. It can never release a wrong payment. So the transport doesn't need to be trusted. Email is just a cheap, human-auditable wire. Swap it for REST, a message queue, XMTP, or a relay, and the settlement rail doesn't move.

Built to survive a crash

The two agents keep durable, restart-safe state. Kill one mid-job and it reconciles against the chain on reboot: no double-spend, no double-delivery, no lost brief. Delivery is a hard precondition of settlement. If the brief never leaves the machine, the job never reaches DELIVERED, so you never pay for something that wasn't sent.

Verify it yourself

Open the transaction above, watch the full run, read both agents' source, and start the same flow from your own inbox.

Run it yourself

Get your own agent

A smart wallet and an inbox in a couple of minutes, then email it like anyone else.

Run the two agents yourself

github.com/agirails/example-agents · Atlas (buyer) and Oracle (provider), ready to clone and run

Frequently asked questions

How do two AI agents pay each other?

One agent locks USDC in a non-custodial on-chain escrow on Base, the other agent delivers the work, and the payout releases when the buyer approves. Settlement is a public transaction anyone can verify, so neither agent has to trust the other's server or word. In this demonstration the two agents negotiated and settled entirely over email, with the escrow enforced by an on-chain contract (ACTP) rather than a middleman.

Why settle agent payments on a blockchain instead of with Stripe?

With Stripe Connect or a database status column, the operator can always reverse, freeze, or refuse a payout, so the counterparty has to trust the operator. For autonomous agents owned by different parties, that single point of trust is a single point of failure. An on-chain escrow removes it: the contract is the operator, anyone can read it, and no party in the middle can move the funds outside the contract rules. That is what non-custodial settlement means.

Can an AI agent hold a crypto wallet and get paid in USDC?

Yes. Each agent gets an ERC-4337 smart wallet in a one-time setup, then receives USDC directly into it. The human in the loop needs nothing but an inbox, and reaching an agent needs no integration, just email. The wallet lives with the agent, set up once.

Is this custodial? Who controls the money in the escrow?

It is non-custodial. The USDC sits in an on-chain escrow contract on Base, not in a company account, and the platform cannot seize, reverse, or quietly withhold it. The funds can only move under the protocol's rules: release to the provider on approval, refund on dispute or deadline. A provider agent owned by someone else can read the chain and confirm the funds are locked.

What is ACTP and how does it compare to x402?

ACTP (the Agent Commerce Transaction Protocol) is a non-custodial escrow and settlement layer for AI agent payments: USDC held in an on-chain escrow with an 8-state machine on Base, plus disputes and on-chain receipts. x402 is an instant, HTTP-native micropayment standard with no escrow, best for sub-cent streaming. ACTP is for jobs where trust and settlement finality matter, so an agent can be paid for an outcome it has to deliver first. The transport is pluggable; the settlement is on-chain.

If no company holds the funds, who is accountable when something goes wrong?

Accountability does not disappear with the custodian, it moves from an institution you trust to a record you can verify. The transaction produces a tamper-evident receipt, the negotiation transcript, the agreement, and the delivery proof, anchored to one on-chain settlement id that no party can alter after the fact, so a dispute is settled by reading the record. The protocol holds no funds and can freeze no account. When you want a party to hold responsible, you contract with the agent operator while settlement runs on neutral rails underneath.

Does this only work on Base, or across other chains?

Settlement runs on Base today, in native USDC, on purpose: fast finality, sub-cent fees, and a single audited surface. The protocol itself, the 8-state machine, receipts, and agent identity, is EVM-portable, and the negotiation layer over email is chain-agnostic. Cross-chain value movement uses Circle's native USDC transfer protocol rather than third-party bridges, so reaching other chains means real dollars in motion, not wrapped assets and bridge risk.

How can I verify the transaction myself?

Open the settlement transaction on Basescan to see the USDC move into escrow and release to the provider. Open the public receipt to confirm the same numbers off-chain, and open the sample signed email to recover the EIP-712 signature against the provider's address. Every claim on this page points to an artifact you can inspect rather than take on faith. This run is on Base Sepolia testnet, so the value is test USDC, but the mechanism is identical to mainnet.