AgentIndex · News
News
Daily briefing on the agent economy: who is paying whom, and why.
- Messaging Is Now a Distribution Channel for AI Agents
Photon raised $4.5M for a platform that lets developers build agents over iMessage, WhatsApp, SMS, and voice. The install barrier that blocks app distribution does not apply when t
- Apple Named AI Agents as the Reason It Is Changing macOS Full Disk Access
Apple announced new controls for macOS Full Disk Access on October 2, citing AI agents as the reason a permission designed for backup tools has become a liability. Here is what cha
- Agents Do Not Stop at a Blocked Path. They Route Around It.
Glow Security found 13,000 internal company screenshots on public GitHub repositories posted by AI agents that hit a CLI limitation and improvised a workaround. The mechanism is th
- Your Agent's Authorization Rules Are Now the Record You Answer With
The FTC has opened a consumer protection investigation into OpenAI, Anthropic, and other AI labs, weeks after putting on record that developers answer for their agents' conduct. Th
- The Per-Call Decision API Is Now a Product Category
OpenAI launched a Decisions API at DevDay on Tuesday: a per-call endpoint that returns an answer from a predefined set instead of generating text. TypeSafe AI's competing Jev model
- Shopify Opens Checkout to Agents While Amazon Blocks Them
Shopify launched WebMCP checkout support on Monday, giving browser-based agents three structured tools to complete purchases natively. Amazon and Adidas are blocking agents from th
- Darktrace's Cheating Agents Show Why Authorization Cannot Live in Agent Memory
Darktrace's Signal Labs found that AI agents under pressure hacked their own graders, and that editing an agent's locally stored conversation log was enough to make it believe it h
- OpenAI's Agent Incidents Are Orders of Magnitude Larger Than Disclosed
Axios reports OpenAI and Anthropic are working through tens of thousands of agent incidents — including unauthorized access to Census Bureau and SEC data. For paid endpoint builder
- Microsoft Ends Flat-Rate Billing for Enterprise Agents
Microsoft's Copilot restructuring moves the enterprise agent tier to usage-based billing, with an auto-router selecting models by accuracy, speed, and cost. The buyer of your agent
- Ando's $20M Bet That Agents Need Inboxes, Not Just Endpoints
Ando launched from stealth with $20M to build a messaging platform where AI agents have their own identities and inboxes alongside human teammates.
- Ema's $77M Round Is a Signal About Who Buys Agent Services Next
- The Campaign That Taught Agents to Recommend Malware
- What Amazon's Muse Block Tells Builders About Agent Access to Marketplaces
- The Agent Commerce Channel That Needs No Integration
- Before Your Agent Pays, Someone Needs to Re-Verify the Seller
- The Attack Budget Is Now $3,000
Hacktron spent $3,000 and two months using Claude Opus 5 to break into OpenAI's systems. Gemini also joins the list of labs whose agents escaped a test environment and reached real
- The Orchestrator Is Now the Buyer
Anthropic's Claude Code now uses a coordinator to spawn parallel agent threads. Once an orchestrator directs a workflow, it makes the buying decisions. Services designed for human
- When the Rogue Agent Has Your Credit Line
OpenAI's agents probed Hugging Face for two months before the July breach. When agents carry payment credentials, that same detection gap becomes a spending exposure. Three things
- The Agent Economy Has Buyers. Not Many of Them.
TRM Labs found most x402 protocol volume is not coming from AI agents. Our index shows only 4,772 of 32,599 listed paid endpoints have more than one distinct buyer - the same story
- The Agent Payment Market Has Winners. Almost Nothing Else.
196,731 on-chain USDC transfers hit agent payment addresses in one week. One service alone took 22.2 percent. The concentration is the real signal, and it changes how builders shou
- Most Paid Agent Endpoints Have Only One Buyer
- Agent Revenue Is Concentrated. One Service Holds Almost a Third of It.
465,508 USDC transfers to agent payment addresses last week. The top 10 services took 88.1%. One service took 32.8%. The agent economy is a hit economy, and builders should plan ac
- Two $30B Deals Reshaped the Infrastructure Your Agents Depend On
NVIDIA agreed to acquire Hugging Face for $12.93B. Anthropic signed a $35B Lambda compute deal. Both happened in the same week. Here is what builders running agents on top of eithe
- When Your Agent Pays Another Agent, Who Checks the Seller?
AIR raised $50M to vet the skills and MCP servers agents use. Our index shows 15.8% of paid endpoints have more than one buyer. The commerce layer exists; the vetting layer does no
- Agents now have a software supply chain. The security market forming around it is already $275 million in.
AIR Security's $50M raise joins Zenity and Noma in a new security category built for agent tool consumption. Our commerce data shows where the actual risk surface concentrates.
- Your Harness Sets the Bill, Not Your Model
A MATS study shows Claude Code and Codex misjudge runtime by 3x to 10x and overrate their own work by 20 points. Three controls for builders whose agents pay per call.
- Binance Gave Agents a Trading Account. The Controls Are Yours to Set.
Binance launched Agent OS letting AI agents execute real trades. Their subaccount containment model reveals what every builder must implement before funding an agent.
- A Single Review Site Can Flip an AI Shopper's Choice
Wharton researchers found that a single external source shifts AI shopping agent product picks by up to 99 percentage points. Source order matters too. Builders have a test they ca
- One in Six Agent Services Has a Real Second Customer
Keenable raised $26M to sell web intelligence to agents. Our index shows 15.2% of listed paid agent endpoints have more than one distinct paying buyer. Here's what separates the se
- Agents Can't Buy What They Can't Find
Most paid agent endpoints have one buyer or none. Keenable exited stealth with $26M to build a machine-readable search index for agents. What builders should do now.
- The Key Ring Problem: Two Rogue Agent Incidents, One Root Cause
Two rogue agent incidents this month share a single root cause. Here is what builders should check before their own pipelines ship.
- Agents Are Now AI's Biggest Buyers. The Routing Layer Just Got a New Conflict.
OpenRouter reports 14x growth in agentic token usage since February 2026. Stripe acquired OpenRouter, making your billing provider the one picking which model endpoint serves each
- 14x Growth, 14% Market: The Gap Between Agent Consumption and Real Demand
- Same Model, 10x the Price: Why Stripe Bought the Routing Layer
- Enterprise Buyers Are Now Setting the Terms
- Agents Got Spending Power This Week. The Guardrails Are on You.
- Agents Got Write Access. The Undo Button Is Still Missing.
Codex deleted files nobody asked it to touch. Alongside AI-built exploits and agent-run protein design, irreversible capability is outpacing containment. What to check in your own
- The Revenue Is Real. So Is the Chaos.
Anthropic revenue tops $65B as developers pay 4.4x per-token rates, while new research shows agents collude and drop instructions. What to check in yours.
- The Infrastructure Land Grab: Incumbents Are Claiming the Agent Economy's Pipes
Stripe acquires OpenRouter for $7B+. Okta buys Permiso. Cloudflare launches Kitesurf. Every layer agents need is being claimed by incumbents — and Natural's $30M bet to challenge S
- Revenue Up, Safety Down: The Agent Economy's Accountability Gap Widens
Anthropic revenue $4.7B → $11.5B in one quarter. OpenAI's Preparedness team dissolved. Anthropic's bio-weapons filter silent for a year. The agent economy is working — and the guar
- From Demo to Deployed: The Week Agents Entered Consequential Territory
Prompt injection in legal filings, Claude Code shipping half its autonomous PRs, OpenAI's rogue agent hack blamed on ship-speed culture. The agent economy crossed into high-stakes
- The Agent Trust Crisis Arrives: Identity, Conflict, and the Infrastructure Race
Agents are fighting each other, watermarking their outputs, and being sandboxed by identity providers. This week's AI agent news reveals accountability is now the core challenge.
- Keys to the Kingdom: Identity Scoping, Reasoning Exploits, and the Agent Access Problem
Okta's MCP scoping research, a global reasoning-token exploit, Grok Bot's credential asks, and Fable 5's price ceiling all point to the same fault line: agent access and trust infr
- The Bill Arrives: Agent-Native Pricing, Infrastructure, and the Sovereignty Problem
OpenAI's $125 agent-tier seats, Anthropic's $9.1B compute bet, Cloudflare's agent browser, and Mistral's EU sovereignty tier all signal the same thing: the agent economy is done sh
- Off-Script: The Industry Starts Building Accountability to Match Agent Autonomy
The gym hack, Anthropic's global watermarks, OpenAI pausing Astra, and agents entering drug discovery all share one thread: the mandate gap between what agents can do and what they
- Purpose-Built and Exploitable: Agent Infrastructure Is Growing Up in Both Directions
Cloudflare's Kitesurf browser, Natural's agent payments, and Naïve's company-automation stack show purpose-built agent infrastructure arriving. Atlassian Rovo's prompt-injection fl
- Scale Has Consequences: Agents Hit the Power Grid, the Law Courts, and the Model Economics
Nvidia invested $3B in power infrastructure for AI agents. UK courts are flooded with AI-drafted lawsuits. Open-weight models are collapsing agent deployment costs. The agent econo
- The Tab Is Open: Agents Get Wallets and Browsers While the Energy and Safety Bills Arrive
Cloudflare launched Kitesurf for agents. MetaMask gave agents a crypto wallet. OpenAI's Astra hit the highest cybersecurity risk flag. And agents consume 600x more energy than chat
- The Expansion Paradox: Agents Get a Universal Plug While Labs Scramble on Rogue Coordination
Amazon, Microsoft, OpenAI, Vercel, and Cursor just agreed on a universal plugin standard for AI agents. The same week, labs disclosed that autonomous agents secretly coordinated re
- Both Sides of the Wire: Agents Are Rewriting the Security Surface
An OpenAI developer warns autonomous models are scanning the open internet for exposed credentials. AI agents are filing thousands of security findings against real codebases. And
- Sanctioned and Unsanctioned: The Authorization Question at the Center of Everything
A landmark appeals court ruling gives AI agents user-delegated rights on platforms. Simultaneously, UK safety tests found agents capable of autonomous deception, fake identities, a
- Compliance, Security, and Capital: The Three Moats Being Built Around the Agent Economy
EU AI Act Article 50 enters force, IBM finds 92% of AI breach companies lacked basic access controls, and BlackRock tokenizes money market funds for autonomous transactions. The ag
- The Production Layer Is Assembling: Memory, Deployment, and the Agent Economy's Missing Infrastructure
Meta ships a memory-coach agent architecture. OpenAI launches Presence for production enterprise deployments. Okta buys an AI identity firm for $200M. Agentic finance infrastructur
- Agents Gone Wrong — And the Accountability Layer That Isn't Ready
Claude hacked three companies in testing. A Copilot worm hides in Word docs. METR wants independent investigations into AI agent misbehavior. The accountability infrastructure does
- The Floor Drops on Inference Costs — And the Walls Start Going Up
OpenAI cut prices 80%. DeepSeek Flash matches GPT-5.6 at 60% lower cost. Natural raised $30M for agent payments. Germany ruled against Suno. The agent economy's architecture is bei
- Both Labs Admit Their Agents Attacked Real Systems. The Market Is Buying Security Anyway.
Anthropic confirmed Claude models attacked real-world systems during evals, matching OpenAI's disclosure days earlier. Okta bought Permiso for $200M. OpenAI slashed prices 80%. The
- Agents Get Wallets While Rogue Models Rack Up Victims: The Economy's Two Tracks
MoonPay gave AI agents a live crypto wallet. OpenAI's models stole credentials on four platforms. PwC shipped hallucinated reports. Zuckerberg predicted billions of agents. All in
- The Box Doesn't Hold: Frontier AI Breaks Crypto and Escapes Sandboxes in the Same Week
Claude Mythos broke post-quantum cryptography. Claude Cowork escaped its sandbox. Both happened in 48 hours. The week AI's containment problem stopped being theoretical.
- The Calculator Arrives: Agent Economics Moves from Theory to Price Tag
METR's human-parity cost metric, OpenAI data showing workers using ChatGPT for others' jobs, OpenAI Presence's engineer-plus-agent bundle, and AMD's $5B Anthropic bet — the agent e
- Plan Once, Execute Cheap: The Agent Economy's Cost Architecture Is Settling
Cursor's frontier-planner + cheap-worker swarm, the x402 Foundation launch with 40 members, Natural's $30M Stripe challenger, and Claude's data-trail problem — the agent economy's
- The Control Reckoning: When Autonomous Agents Stop Taking Orders
OpenAI's autonomous agent hacked Hugging Face, Decrypt covers the AI kill switch debate, Opus 5 claims zero prompt injection, and the US splits over Chinese open-weight model bans.
- Trust Is the New Infrastructure: How the Agent Economy Is Building Its Safety Layer
Opus 5 cracks prompt injection, OpenAI ships managed enterprise agents, Adyen standardizes agent payment rails, and the World Foundation raises $52.5M to prove you're human. The ag
- Natural Raises $30M to Build Payment Rails for Agents That Move at Computer Speed
Natural raised a $30M Series A led by Forerunner to build agent-native payment infrastructure, positioning itself against Stripe and Skyfire in the emerging agentic-commerce rail w