Every AI company has been vague about exactly where its training data came from. This week, a court forced one of them to put a number on it.

Anthropic will pay $1.5 billion to authors and publishers, the largest copyright recovery in history, after building part of its training library from pirated books. We'll break down the line the judge drew between buying books and stealing them, and what it means for every other AI company watching nervously.

Plus a tech company that cut a fifth of its staff and blames a "reset" instead of AI, a hiring trend in accounting and banking that's outpacing tech itself, and a math problem a chatbot casually solved during the World Cup final.

Let's get into it.

The best assistants don't multitask their attention across a hundred tools. Neither does Catch. It's an AI agent that focuses on one thing — the admin work you'd rather not touch — and does it exceptionally well.

Scheduling, flights, restaurants, follow-ups, vendors, clients. You hand it over; Catch handles the back-and-forth and comes back with it done.

No context-switching. No dropped balls. Just your admin, quietly cleared — so your focus stays on the work only you can do.

Meet the agent built for admin, and it'll be ready to work before your next meeting.

Get started at catchagent.ai — and give your attention back to what matters.

Start now for free

A federal judge approved Anthropic's $1.5 billion settlement with authors and publishers this week, the largest copyright recovery in history. Anthropic will pay roughly $3,000 per book across more than 482,000 titles, and 91% of eligible books have already been claimed.

The case turned on how Anthropic built its training library, not on whether it trained Claude on books at all. The court ruled that training AI on books is fair use when the books were legally bought, but Anthropic's separate library of 7 million books downloaded from pirate sites like Library Genesis crossed the line.

The case started in 2024, when novelist Andrea Bartz and two other authors sued Anthropic over exactly that pirated library. Buying the books and training on them turned out to be legal; building the dataset off piracy wasn't, no matter how good the resulting model got.

Every major AI lab has similar lawsuits pending, and this ruling just handed future plaintiffs a template and a real dollar figure to point to. Expect more settlements like this one within the year, and a higher price tag industry-wide on training data that's actually clean.

Why it matters: Every AI lab just watched a court attach a real number to sloppy data sourcing, so labs and publishers both need people who understand data licensing and rights clearance, a function that barely existed two years ago. If you're anywhere near AI strategy, product, or ops, this week is a good time to find out how your own company sources the data it trains or fine-tunes on, since "we know exactly where this came from" just became a real competitive answer instead of a legal footnote.

Accounting offices grew job postings requiring AI skills by 55% year over year, and commercial banks grew theirs by 51%, both outpacing most pure tech industries, according to the Bipartisan Policy Center's AI Skills Dashboard. The specific skills showing up are telling: Copilot and prompt engineering in accounting firms, Hugging Face and large language model development inside banks.

That tracks with tech hiring more broadly. Dice's 2026 Tech Jobs Report, based on 7 million postings, found "agentic AI" is the single fastest-growing skill requirement in tech, up 587% year over year, from a near-zero base twelve months ago.

Put those together and the takeaway isn't "learn to code," it's "learn to run the AI that already exists in your own industry." Accounting, banking, and consulting firms aren't only cutting headcount with AI, they're hiring people who can make it work inside the business they already have.

Why it matters: If you're weighing whether to leave consulting or professional services for a tech job to be "closer to AI," this data suggests the AI jobs are coming to you instead. Look at whether your own firm already has a Copilot rollout, an AI pilot, or a prompt-engineering initiative sitting unclaimed, and volunteer to help run it this quarter, since being the person who knows both the business and the tool is exactly the combination this week's data says gets paid the most.

→ Land your next role faster with Tailr, our Claude-powered job search and networking tool: Start searching

→ If you're targeting $150K+ roles, have our team run your entire job search for you: Apply here

→ Build your own personal AI workforce handling your work, business, and personal life: Access to ready-made agents

Build a real comp benchmark before your next negotiation, not a guess. Copy this into Claude or ChatGPT: "I'm preparing to negotiate compensation for a [job title] role at [company or company type], based in [location], with [X years] of experience. Using publicly known compensation data patterns (Levels.fyi-style ranges, industry norms), help me build a realistic total comp range for this role and level, broken into base, bonus, and equity if applicable.

Then help me draft three specific talking points I can use to justify landing at the top of that range based on my background. Keep it concrete, not generic negotiation advice.

Quick status check on two predictions still in flight: Fireworks' bet on tripling headcount to 600 employees by December has no result yet, and Volkswagen's job-cut plan actually got bigger since we last flagged it. An internal memo now models up to 140,000 cuts, up from the 100,000 we mentioned Monday, with works council meetings set for August 25 and 26.

New prediction: now that Anthropic's settlement is approved, will another major AI lab (OpenAI, Meta, Google) reach a similar copyright settlement within the next 6 months?