Why the question of whether to license AI is harder than it looks

Hi, I’m media innovation journalist Ulrike Langer and you’re reading News Machines. If you get value out of my newsletter week after week, please consider upgrading to the premium version. And if you can’t, the easiest way to support my work is by clicking on the ad down below in the post. Every click is worth more than $2.

A publisher in 2026 faces three open questions about AI: 

Two events a few weeks ago show why these three questions are connected.

On April 29, Sam Altman went on Nick Thompson's Atlantic podcast and said AI agents should pay publishers. By mid-May, the clip was making the rounds. He gave numbers. An agent pays 17 cents to read an article. A human pays a dollar to read the whole thing. The publisher, in Altman's example, names the price.

On May 7, Amazon Web Services launched Bedrock AgentCore Payments — a real product, with documentation, APIs, and a pricing surface that already runs. AWS built it with the payment systems Coinbase and Stripe. Agents can register a payment source, take on a spending limit, and conduct transactions autonomously. AWS lists financial analysis with paywalled content as one of the five canonical use cases: agents paying for access to “real-time market data behind paywalls.”

One is a CEO talking on a podcast. The other is a product that shipped. And when you read about what shipped, it does not do what Altman described.

Read the AgentCore documentation and you discover that the architecture is split in two.

AgentCore handles the agent side of payment. The agent's operator sets a budget, the agent transacts within it, AWS handles the rest. A few lines of code, two payment partners, and an agent can pay for any paid resource on the open internet.

The publisher side is a different story.

For an agent to pay a publisher, the publisher has to expose content through an x402 endpoint — an open protocol Coinbase developed that lets a website respond to an agent's request with “payment required” plus a price, then deliver the content once payment is verified. The publisher sets the price. AgentCore just lets agents pay it.

A publisher who has built an x402 endpoint can name a price an agent will encounter. The Coinbase x402 Bazaar, which AgentCore connects to directly, already lists more than 10,000 paid endpoints. Almost none belong to news publishers.

This is what Sam Altman's micropayment scenario quietly assumes. AgentCore solved the easier half. The harder half is the publisher's problem.

You may think this doesn't matter because AgentCore just launched. But infrastructure decisions made now will be hard to reverse. If AgentCore becomes the dominant agent-payment rail — and AWS has the distribution to make that happen — the architectural choice baked in today is the one that publishers will live with for years.

Now the other story, because it is connected.

At Newsrewired in May, Ezra Eeman, strategy and innovation director at the Dutch public broadcaster NPO, said it bluntly: “Nobody knows the right model.” Crawl for free vs. pay per crawl, links and attribution vs. exclusive deals, revenue share somewhere in the middle — Eeman lays out the options and concludes the economics have not yet worked out.

He is right. But three coalitions of publishers have already committed to three of those options, and the options are not compatible with each other. One is built around pay-per-interaction. One is built around blanket licensing covering everything for a flat rate. One is built around publishers collectively owning the distribution layer agents pay to access. These three models compete for the same publisher revenue stream. They do not stack. Whichever one becomes the default makes the other two marginal.

The first is SPUR. In February, the BBC, the Financial Times, the Guardian, Mediahuis, Sky News, and the Telegraph founded SPUR — the Standards for Publisher Usage Rights coalition. They announced it as a collective response to AI. The founding CEOs signed an open letter. The trade press covered it as “the British publishers' NATO for News.”

A month after SPUR's founding, the News/Media Alliance — representing 2,200 US news and magazine outlets — signed a licensing deal with Bria, an AI company that pays publishers per use. Payment is based on how often content surfaces in enterprise AI tools, such as financial copilots, legal chatbots, and internal search.

In May, WAN-IFRA reported that dpa, the German press agency owned by 174 German media companies, is repositioning itself as a “trusted info layer” for AI agents.

SPUR is built around shared standards for a transactional future. Its own framing names pay-per-crawl and pay-per-inference as the structures worth working out — not blanket licensing, not infrastructure access.

The Bria deal — and the older Danish Press Publications' Collective Management Organization, which 99% of Danish news outlets joined in 2021 — operates on the blanket-licensing model. Negotiate a rate. Cover everything. Let AI companies use the content under the terms of the deal without metering every interaction.

dpa points somewhere else again. AI agents don't license dpa's content. They pay for access to a verified-content distribution layer publishers own collectively. The unit of payment is access to the layer, not the article.

A publisher joining one of these coalitions is working against the assumptions of the other two.

The early evidence is mixed. AgentCore points transactional, which favors SPUR's model. The Meta-News Corp deal at up to $50 million per year for three years points blanket-licensing. Microsoft's Publisher Content Marketplace has been described in language that could fit any of the three. We do not know yet which one will win.

So when someone tells you that publishers need to join a coalition, they are skipping a question. Which coalition? Built for which future?

SPUR's founding members include five of the most powerful news organizations in the English-speaking world: the BBC, the Financial Times, the Guardian, Sky News, and the Telegraph, alongside Belgian-Dutch group Mediahuis. The BBC operates under a Royal Charter with annual public funding measured in billions of pounds. The FT is owned by Nikkei and has direct relationships with every major technology platform. The Guardian has the Scott Trust behind it. None of these publishers needs a coalition to get a returned phone call from Microsoft or Amazon or OpenAI.

They built one anyway, which is rational — collective leverage beats individual leverage even when individual leverage is already very high. But notice the pattern. Coalitions in the AI moment are forming around publishers with already strong individual positions, or in markets with pre-existing collective infrastructure.

Denmark is the inverse case. The CMO formed in 2021 because Danish copyright law and a long collective-rights tradition gave Danish publishers an existing structure to plug into. dpa is similar — a 1949 cooperative whose 174 shareholder companies include most of Germany's regional newspapers, broadcasters, and magazine publishers.

Publishers without a coalition seat in May 2026 are in markets without a collective-rights tradition to repurpose for the AI moment. A mid-sized US regional publisher cannot join SPUR. There is no US equivalent — the News/Media Alliance offers opt-in deal frameworks but is not a standards body. A mid-sized publisher in most Asian markets, in most of Eastern Europe, in much of Latin America has nothing comparable to dpa or the Danish CMO to fall back on. The privilege divide is not between large and small publishers. It is between markets that built collective publisher infrastructure decades ago and markets that did not.

This has a useful flip side. If you are a German regional publisher, you sit inside the coalition that is building dpa-iq, the trusted info layer. You have a stake in what gets built. The question is what you want the your coalition to build for you — and whether the dpa-iq architecture matches the future you think is arriving.

If you are not in a market with a national collective-rights body, the next section is for you.

Whatever coalition you can or cannot reach, there is one move every publisher can make directly. Build your own house.

The Economist is doing this with its two-track web — restructuring content outside the paywall for AI answer engines. Belgian-Dutch publisher Mediahuis runs a multi-agent pipeline for first-line news, where AI agents handle writing, fact-checking, legal review, and image selection before a human editor publishes. It is acting at both the coalition layer (it is a SPUR founding member) and the individual-build layer. DMG Media, parent of the Daily Mail, built Mail iQ, a multi-agent system that produces more than 300 social media assets a day. The Philadelphia Inquirer's engineering team built an internal agent that, when a Jira ticket is flagged “readyforai,” fetches the spec, pulls documentation from Confluence, grabs designs from Figma, creates a git branch, and writes the code using Claude Code.

Building your own house comes down to three things: a structured archive of your content, a way for AI agents to query it directly, and a way to charge them when they do. That is what Altman, the Economist, dpa, and the entire coalition layer assume already exists somewhere.

Here is why this matters: Building your own house is the only move that pays off regardless of which coalition wins. If the transactional future arrives, an MCP server in front of your content lake lets you transact. If the blanket-licensing future arrives, a clean content lake with structured licensing metadata makes you ready for a deal. If dpa's infrastructure-access future arrives, you have the structured content to plug into the layer.

The three questions a publisher faces in 2026 — how to get paid, which coalition to join, what to build internally — have a right order. Build first. Then choose a coalition that matches what you have built. Then sign deals that match both.

A publisher running the sequence the other way is signing deals before knowing what has been built, joining coalitions before knowing which architecture wins, and hoping it works out. It might. But hope is not a strategy you can execute.