This week added three separate developments that point in the same direction: AI content is getting more legally complicated, and proof of what you made and when is becoming the thing that protects you. A record settlement closed. A new EU disclosure law took effect, with real fines attached. And a jury trial opened that could shape how AI image generators are treated under copyright law going forward. None of it is abstract. Below is a checklist to work through this month, not a summary to read once and forget.
The week in three lines
A federal court gave final approval to the 1.5 billion dollar settlement in Bartz v. Anthropic on 20 July 2026, the largest copyright settlement of its kind and a plain signal of what unauthorized training data can cost a company. On 2 August 2026, Article 50 of the EU AI Act came into force, requiring disclosure when content shown to users is AI-generated or AI-manipulated, with fines that can reach EUR 15 million or 3 percent of global turnover. And this month, a jury trial opened in Andersen v. Stability AI, another test of how image generators handle the training data question. No verdict has been reached yet.
Your September checklist
Read through this list once, honestly. If you cannot answer a question with a clear yes, that is the item to fix first.
- Do you know whether your AI-generated content needs an Article 50 disclosure? If your business shows EU users any content that was generated or materially edited by AI, a label is very likely required now that Article 50 is in force. Check this before your next campaign publishes, not after.
- Do you have dated, independent proof of your own creative or document work? Proof needs to exist before a dispute starts, not once one is already underway. Being able to show exactly when you created something matters more than being able to argue about it after the fact.
- Are you tracking which AI tools your team actually uses, and what those tools were trained on? Every recent lawsuit and settlement traces back to a training data question. Know which vendors your team relies on day to day, and whether those vendors have disclosed their training sources.
- Have you reviewed vendor contracts for indemnification language? If a training data claim ever reaches your business through a tool you pay for, find out now, not during a legal filing, whether that vendor covers you or leaves you exposed.
- Does your team understand the difference between using AI output and training on someone else's protected work? That distinction sits at the center of the cases moving through court right now, and it changes what kind of risk your business is actually carrying.
- Do you have a written policy for AI disclosure on anything public-facing? A one-page internal policy, agreed before enforcement questions arrive, is far cheaper than sorting it out under pressure.
- Have you set a reminder to revisit this list next month? The rules in this space are still moving. A short monthly check is a low effort way to keep your business current instead of catching up later.
None of the above needs to be complicated, but every item does need an actual answer, not an assumption. If Article 50 disclosure is the one you are least sure about, start there. We put together a step by step guide covering what counts as AI-generated content under the regulation, how to write a disclosure that meets the requirement, and where it needs to appear. Get the Article 50 checklist and work through it this week, before it turns into next month's problem.
