Fetching from the wire…
Public story · 2026-07-21 · high
Over 170 publishers now get paid per token an agent reads, the first structural fix for content owners who saw zero revenue from AI scraping.
Why now: The marketplace shows up in coverage dated July 21, 2026, as publishers keep looking for a way to charge AI agents for the content they consume.
Bigdata.com launched a marketplace that charges AI agents by the token for content from more than 170 publishers, per aiagentstore.ai.
Publishers have had no clean way to get paid when an AI agent reads and reuses their work. This marketplace ties payment directly to token consumption, not seats or flat API fees, across a network RavenPack puts at over 170 premium providers.
RavenPack also claims the platform cuts token consumption up to 100x compared to naive retrieval. That's meant to let an agent pull far less raw text for the same answer. That number comes from RavenPack, not an independent test, and aiagentstore.ai's report doesn't say how it was measured or against what baseline.
The pricing model is the real story here, not the compression claim. Seat-based and flat-fee licensing never mapped to how agents actually consume content: some queries touch a paragraph, others ingest a whole archive. Metering by token lets a price track actual use. It's the same problem every publisher deal with an AI company has been stuck on since agents started reading the web instead of humans.
If the 170-plus number holds and other providers see real payouts, expect competing metered marketplaces within a year. If the token math is the only compression on offer, and payouts stay thin, publishers will read this as another data license with better marketing.
Each link below shares sources, entities, or timing with this story.
The comparison is against GB300 NVL72, with 35x lower cost per million tokens, measured on the SemiAnalysis AgentX benchmark using real recorded agentic coding sessions with context growth, tool calls and sub-agent spawning preserved (NVIDIA). DeepSeek V4 Pro and Qwen3.5 were...
Agent Intent-Based Access Control compares an agent's reasoning, tool calls and data access against its originally assigned task with adjustable sensitivity, flagging or blocking anything outside scope. It evaluates entire sessions rather than individual requests, specifically...
The August 3 piece defines agentic BI explicitly against dashboards: a dashboard is a fixed set of charts built ahead of time by someone guessing which questions would matter, while an agent interprets the question in the moment, decides which tables to query, and can trigger...
Microsoft's July 23 release targets a genuine gap: harness-based agents like Claude Code and Codex drive multi-turn reasoning, tool use, and external system access but were hard to train end-to-end with standard open RL infrastructure. The trick is decoupling training from inf...
Raffi Khatchadourian's replay benchmark measures behavioral instability through three channels that need no access to hidden reasoning text: tool-call trajectories, evidence contacts, decision concentration (arXiv 2607.20491). Across 8,127 replay episodes over 10 models and 3...
Self-hosted agents read and write their own memory and config to function, which means an attacker can compromise one entirely through legitimate OS system calls with no exploit involved (arXiv 2607.17986). The paper builds a 23-cell attack matrix across Target, Mechanism, Gra...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.