The $1.5B Settlement That Isn't Really a Win

ANTHROPIC'S $3,000 APOLOGY

Authors Get Paid, But the Math Tells a Different Story

The Rundown: The 'largest copyright recovery in history' works out to about 2% of what authors could have claimed under federal law.

The details:

• Anthropic settled for $1.5 billion after downloading roughly 500,000 copyrighted books from pirate sites like Library Genesis.
• Authors receive approximately $3,000 per pirated book — federal law allows up to $150,000 per work for willful infringement.
• The fine represents less than 0.2% of Anthropic's $965 billion valuation.
• 91% of the roughly 440,000 eligible claimants have already accepted the settlement.
• The ruling doesn't require AI companies to get permission to train on copyrighted work — it just says they need a legal copy.

Why it matters for us: If you're a community publisher sitting on decades of archived content, this settlement confirms AI companies need legal copies to train on — but it doesn't establish that you're owed anything for the training itself. The licensing question is still wide open.

Read at creativebloq.com →

THE AUTHOR'S DILEMMA

Lead Plaintiff Says He's 'Much Poorer' After Winning

The Rundown: Charles Graeber spent two years fighting Anthropic. He's getting $6,200 for two books. He passed up paying work to do it.

The details:

• Graeber, author of The Good Nurse (adapted by Netflix), was one of three lead plaintiffs in the class action.
• He'll receive roughly $3,100 per book for the two works Anthropic pirated.
• The lawsuit required two-plus years of travel, legal discussions, and turning down other assignments.
• Graeber called the outcome proof that 'piracy is a crime' but stopped short of calling it a win.
• More than 300,000 writers are part of the settlement class.

Why it matters for us: For a weekly paper considering joining a class action against an AI company, this is the reality check: even winning costs time and money you may not have. The payout won't cover the opportunity cost.

Read at npr.org →

AI NEWS GETS IT WRONG

Prism News Published a Fake Flood Death in New Mexico

The Rundown: An AI-only news site reported a resident died in recent flooding. The death was real — but it happened in a different year, in a different flood.

The details:

• Prism News says all its stories are written and published using AI without human input.
• The outlet reported a flood death in Gallup, NM that actually occurred in a previous year's separate flood.
• The error was caught by Source New Mexico, a human-staffed outlet.
• The Gallup Independent closed in February after a century of coverage, leaving a weekly paper and monthly magazine.
• 1,524 U.S. counties now have only one remaining local news source, usually a weekly newspaper.

Why it matters for us: This is exactly what happens when AI fills a news desert without human oversight. For community publishers still staffing their beats, this is your competitive moat — you know when someone actually died and when they didn't.

Read at san.com →

AUSTRALIA PUSHES FOR AI FEES

Publishers Want AI Companies to Pay Like Google and Meta Did

The Rundown: Australian news organizations are lobbying for AI licensing fees, hoping to avoid another years-long bargaining code fight.

The details:

• Focus is shifting from social media bargaining codes to AI companies using news content without payment.
• Publishers want the government to move faster than it did on the original news bargaining code.
• LinkedIn has added a feature for users to flag AI-generated 'slop' posts.
• The push parallels the successful effort that got Google and Meta to pay Australian publishers.

Why it matters for us: If Australia establishes AI licensing fees for news content, that framework could spread. For a U.S. community publisher, this is worth tracking — any precedent that says AI companies owe you for training on your archives changes your revenue picture.

Read at afr.com →

THE UPDATE

Quick Hits From the Week

The Rundown: Smaller items that crossed the desk.

The details:

• LinkedIn's new 'AI slop' flagging button is live — users can now report generated posts cluttering their feeds, which may affect how AI content surfaces on the platform publishers use for audience building.
• The Anthropic settlement doesn't release the company from future claims based on what Claude actually generates — the output question remains unresolved.
• Since 2005, nearly 3,500 newspapers have closed nationwide, including 136 in the last year alone, per Northwestern's 2025 State of Local News report.

What is?

Web Scraping

What it is: Web scraping is using software to copy content off websites automatically, page after page, far faster than any person could. A program visits a site, grabs the text, images, or data, and saves it. It's how search engines index the web, how price-trackers watch competitors, and, most relevant here, how AI companies gathered much of the material they trained their models on.

Why publishers care: Your archive is a scraping target. A lot of what these models “learned” they learned by scraping sites like yours, usually without asking. That's the quiet engine behind every copyright fight in this issue. The good news: you can see it in your server logs, and you can set rules (a robots.txt file, or a block on known AI crawlers) that tell the bots to stay out. It won't stop everyone, but it turns “they took it” into “they ignored a posted no.”