The AI Licensing Gap Is Already Here

THE LICENSING LOCKOUT

Brookings finds local publishers effectively absent from AI deals

The Rundown: A new Brookings analysis maps the AI content licensing market and finds it's replicating the same power imbalances that damaged journalism during the search era.

The details:

• Click-through rates from AI systems collapsed six-fold by Q4 2025, eroding even the referral benefits large publishers gained from early licensing deals.
• Local newspapers, regional broadcasters, and ethnic media are effectively absent from the AI licensing market despite providing critical civic journalism.
• Current valuation frameworks focus narrowly on referral traffic lost, ignoring publisher contributions including training data, factual grounding, and civic legitimacy.
• The European Parliament and UK House of Lords both concluded that statutory licensing frameworks with mandatory transparency are needed.

Why it matters for us: If you're running a 10,000-circ weekly, you're currently locked out of this revenue stream entirely. The report suggests collective bargaining through state press associations may be the only path to the table — start that conversation with your association this month.

Read at brookings.edu →

MICROSOFT'S FIRST AUSTRALIAN DEAL

Nine Media gets paid for journalism powering Copilot

The Rundown: Microsoft struck its first Australian media licensing agreement, paying Nine Media for content to boost its Copilot AI assistant.

The details:

• The deal marks Microsoft's first media payment agreement in Australia, establishing a precedent for the region.
• Nine Media content will be used to enhance Copilot's responses on Australian news and information.
• The agreement follows similar deals Microsoft and OpenAI have signed with large publishers in the US and Europe.
• Smaller Australian publishers are not included in this deal and face the same access barriers as US local outlets.

Why it matters for us: These deals keep going to national players with legal teams and larger circulations. For a regional daily or community weekly, the question isn't whether your content has value — it's whether you can get anyone at Microsoft to return your call.

Read at patrickwhite632.substack.com →

THE SANDBOX APPROACH

Norwegian publisher Amedia built an AI testing environment in a month

The Rundown: A Reuters Institute survey shows 90% of newsrooms now use AI in production, but only one-third have an actual AI strategy — Amedia offers a template for the rest.

The details:

• Amedia's AI sandbox took just 1-2 developers about a month to build, creating a secure environment for journalists to experiment with generative AI.
• The sandbox keeps paywalled content out of public LLMs while allowing staff to test transcription, draft generation, and research tasks.
• Back-end automation like transcription and copyediting is the top AI priority for 56% of newsroom leaders surveyed.
• Only one-third of newsrooms have a formal AI strategy despite 90% already using AI tools in production.

Why it matters for us: You don't need a tech team to start. A 10-15 hour investment from a developer — or a technically-inclined staffer with ChatGPT Team access — can create a space where reporters experiment without accidentally feeding your subscriber content into the training pool.

Read at innovation.media →

USA TODAY'S PHASED ROLLOUT

Gannett's flagship shows how to introduce AI without a revolt

The Rundown: USA Today partnered with Microsoft to bring AI into daily workflows, starting with low-risk tasks and letting reporters build trust before expanding.

The details:

• The implementation began with research assistance and background information gathering — tasks where errors are caught before publication.
• Content summarization tools help journalists process large volumes of meeting minutes, reports, and documents quickly.
• Editorial oversight remains central throughout — AI assists but makes no publishing decisions.
• Workflow automation handles repetitive tasks like formatting and metadata tagging, freeing reporters for original work.

Why it matters for us: The lesson for a community paper: start with summarizing city council packets or pulling background on candidates. Let reporters see the time savings on low-stakes work. The trust builds from there.

COPY EDITORS OUT, AI IN

French publication Gazette des Communes cuts human editors — Le Monde calls it a quality threat

The Rundown: Gazette des Communes replaced human copy editors with AI, and Le Monde's coverage suggests the results aren't pretty.

The details:

• The French publication moved to AI-based copy editing as a cost-cutting measure.
• Le Monde argues the shift threatens the title's editorial quality and reader trust.
• The case is being watched across European publishing as a test of how far AI can replace editorial judgment.
• No U.S. community newspapers have publicly announced similar full replacements of copy desk staff.

Why it matters for us: There's a difference between using AI to catch typos before your copy editor sees the proof and using AI to replace your copy editor entirely. The first saves time. The second will eventually cost you credibility — and readers notice.

Read at patrickwhite632.substack.com →

ANTHROPIC VS. THE WHITE HOUSE

Export controls forced Claude Fable 5 offline for 18 days

The Rundown: Anthropic's newest models went dark worldwide after U.S. export controls kicked in — then came back with new safeguards and a framework for the industry.

The details:

• US export controls blocked access to Fable 5 and Mythos 5 from June 12-30 after Amazon researchers found a method to bypass safeguards.
• Anthropic suspended access to all users globally because they couldn't verify nationality in real-time.
• A new industry framework for scoring jailbreak severity is being developed with Amazon, Microsoft, Google, and other partners.
• The reported jailbreak was classified as minor — the same vulnerabilities existed in older Claude versions and competitor models.

Why it matters for us: If your newsroom relies on Claude for daily workflows, you just learned these tools can vanish overnight based on policy decisions in Washington. Build redundancy. Know your backup.

Read at anthropic.com →

What is?

Loop Engineering

What it is: For a couple of years, the trick to getting good work out of AI was "prompt engineering" — wording your request just right, like dictating a letter word-for-word, so the one answer that came back was usable. Loop engineering is the newer idea quietly replacing it. Instead of crafting one perfect request, you give the AI a goal and a way to check its own work, then let it run in a loop: it takes a crack at the task, looks at what it produced, fixes what's off, and tries again — over and over — until the result meets the mark. You're no longer hunting for the perfect sentence. You're setting up the routine and letting the machine grind toward "done."

Why publishers care: It changes what "being good at AI" even means — and that's good news if you were never going to become a prompt wizard. The skill is shifting from clever wording to plain judgment: knowing what "done right" looks like and telling the machine how to check itself against it. You already have that part — you know a clean story, a fair headline, a house-style caption when you see one. It also explains why AI is suddenly finishing bigger jobs than it could a year ago: a tool that re-checks and re-tries its own work needs far less hand-holding than one you had to word perfectly on the first try. Watch for it in the tools you buy — the good ones increasingly work this way under the hood.