Today's briefing
Where Your AI Got Its Training Data Is Now Your Problem
Sony and Warner take Anthropic to court over pirated music. The lesson for small businesses is about which plan you bought, not who wins.
The one that matters
Two of the world's biggest music publishers sue Anthropic over how Claude was trained
What happened
Sony Music and Warner Chappell — between them owners of a very large share of the world's song copyrights — have sued Anthropic, the company behind the Claude AI assistant. The claim is not mainly about what Claude writes back to you. It is about where the training material came from. The labels allege a broad campaign of intellectual property theft, including downloading music from torrent sites, the same file-sharing networks record companies have been fighting for twenty years. The damages sought run into billions of dollars. This is wider than the earlier argument about Claude reciting song lyrics.
What it actually means for your business
The important shift is where the legal risk sits. For two years the fight has been about outputs — does the AI copy someone's work when it answers? This case attacks the input: how the material was obtained in the first place. That is a much harder thing for an AI company to explain away, which is why cases of this shape tend to end in large settlements rather than clean legal answers.
For you as a customer, the practical consequence is boring and specific: your exposure is in what you publish, and in which plan your staff are using. A 600-student school generating fee-reminder letters and timetable drafts has essentially no copyright risk. The same school producing an annual-day video with AI-generated music, or a 12-person marketing agency shipping AI-written copy and images to a client, is in different territory. That material goes into the world under your name, and it is your contract with your client — not Anthropic's contract with you — that decides who pays if a complaint arrives.
Business and enterprise tiers from the major AI vendors generally carry some form of copyright indemnity. Free and personal accounts generally do not. If some of your staff are doing client work on free consumer logins — and in most small firms, some of them are — you are carrying the risk without the cover. Closing that gap costs roughly ₹1,700–2,500 per user per month in India, or CAD $30–35 in Canada. For a ten-person agency that is real money, but it is not a hard decision.
What this does not mean
- It does not mean Claude or any rival assistant is about to be switched off. Music publishers want to be paid, not to remove the product.
- It does not mean work you generated last year is now unlawful to use. This is a US suit against a US company. Neither Indian nor Canadian courts have ruled on any of it.
- It does not mean you need an AI compliance audit. Consultants are already selling those off headlines like this one. Below about 200 staff, that is a paid solution to a problem you can handle yourself in an afternoon.
- Equally, do not accept a vendor telling you their indemnity leaves you fully protected. Read the liability cap and the exclusions. They typically exclude the case where someone on your team asked the tool to imitate a named artist, brand or style.
What a sensible owner does this month
- Spend thirty minutes listing every place AI-generated words, images, music or voice reach a customer. Most owners find two or three they did not know about.
- Move anyone doing client-facing creative work off free accounts and onto a paid business plan, then actually read what the indemnity covers.
- Add one line to your client contracts covering who owns and who warrants AI-assisted deliverables. Your lawyer will charge very little for this.
- Do not change AI vendors because of a lawsuit. Every major vendor has one.
Also worth knowing
-
Attackers are stealing logged-in AI sessions to take over accounts
Anthropic has warned that attackers are hijacking active Claude sessions rather than cracking passwords, which means a strong password alone does not protect the account. If your staff use AI tools on shared or personal machines, turn on two-factor authentication and revoke sessions when someone leaves. This is the same housekeeping you already owe your email and accounting logins.
Search Engine Journal ↗ -
AI that operates your web browser is now open to everyone, including without asking permission
Anthropic has made its browser-operating AI generally available, with a mode that lets it carry out steps without pausing for approval. For a business, the risk is not the technology but the browser it runs in: never let it loose in a profile logged into your bank, payroll, or GST portal. Give it a separate browser profile with only the accounts it genuinely needs.
finance.biggo.com ↗ -
Cyber insurers are rewriting policies for AI agents that go wrong
Insurers are adjusting cover as businesses hand real tasks to AI agents that can act on their own. If you carry cyber insurance, one phone call to your broker this week is worth making: ask whether a loss caused by software acting on your behalf is covered or excluded. The answer is likely to change your appetite for letting agents touch money or customer data.
Lee News Central ↗ -
Anthropic trims weekly usage limits on its coding tool by 17%
The weekly cap on Claude Code is being cut, which matters only if you or your development contractor pay for it. If a vendor bills you for AI-assisted development, expect the subject of top-up costs to come up. It is a reminder that AI subscription pricing is still unsettled and should not be treated as a fixed line in next year's budget.
BleepingComputer ↗ -
The most-read argument among engineers this week: culture beats AI on productivity
A widely discussed piece argues that the biggest gains in output come from how a team is managed, not from what tools it buys. Worth ten minutes before you sign an AI contract sold on productivity numbers. If your team is slow because of unclear priorities or too many approvals, no assistant fixes that.
Hacker News ↗ -
Some teams are trying one AI-free day a week
The practice of setting aside a fixed day where staff work without AI assistance drew a lot of discussion, the argument being that constant use erodes the judgment needed to check the output. It costs nothing to trial for a month. If your team cannot function on that day, you have learned something useful about how much you are trusting unverified output.
Hacker News ↗ -
The Philippines is putting Microsoft's Copilot into its public schools
A national education department has partnered with Microsoft to bring AI assistance into public schools, a pattern school operators in India and Canada will see arrive locally. If you run a school, the useful question is not which tool but what teacher training and support come bundled. Rollouts of this kind fail on training and habit, not on licences.
YugaTech ↗ -
Caterpillar is applying decades of mining automation lessons to AI rollouts
A company that spent years making autonomous machines work at remote mine sites is now applying that experience to deploying AI internally. The transferable lesson for a factory or workshop owner is staged rollout on narrow, well-defined tasks before anything broader. Most of the difficulty sits in process and people, which is exactly the part vendors do not sell you.
TechCrunch AI ↗
How this briefing is put together
Every morning we read the day's AI announcements and reporting from the companies themselves and from the technology press, then pick the handful that actually change something for a working business. The analysis is ours and it is written for owners and managers, not engineers. Every story links to its original source above — read them, and disagree with us where we've got it wrong.