What happened
AI tools like ChatGPT can write and answer questions because the companies behind them first had them read huge amounts of text from the web. That process is called training. A good share of that text is newspaper journalism, and several publishers, arguing this infringes their copyright, have taken the AI companies to court. In the first week of September, both the US government and the EU made moves on the issue.
In the US, The New York Times is suing OpenAI (the company behind ChatGPT) and Microsoft for copyright infringement. On 1 September, the Department of Justice filed a written brief in that case, asking the judge to side with the two companies. Its reasoning: when AI reads articles, what it learns are the patterns in how words go together, and what it then produces is different from the original reporting, so the use counts as “fair use”, which the law permits, rather than infringement. Barring AI training on copyrighted material, it added, could “severely hamper” technological progress.
The brief isn’t a ruling, and the judge doesn’t have to follow it. Still, US press coverage described this as the first time the US government has formally taken a position in the current wave of AI copyright cases.
Later that week, on 4 September, two more US papers, The Seattle Times and Newsday, sued OpenAI and Microsoft. Besides copyright, they added a claim of trademark dilution, which in plain terms means damage to a paper’s good name: the two papers allege that AI makes up content that doesn’t exist and says they wrote it.
Europe is going about it differently, through a law written specifically for AI, the AI Act. In early September, the EU body that oversees AI (the AI Office) sent its first batch of information requests to more than 30 AI model makers, the companies that build AI like ChatGPT, and one of the main areas the questions cover is copyright. The EU hasn’t said who received them.
The EU can do this because the AI Act’s rules aimed specifically at these AI model makers became enforceable on 2 August: the companies have to account for the content they trained their AI on, and the AI Office can request information from them and, where needed, impose fines.
Why it matters to you
What Washington and Brussels are dealing with comes down to the same thing: whose articles AI companies took to train their AI, and how. For anyone running a business, though, something else hits closer to home: when customers ask AI about you, how does it describe you, and does it get you wrong?
Say your bakery moved two years ago, but an old food-blog post with your former address is still online. A customer asks ChatGPT where to find you, the AI answers from that old post, and the customer turns up at an empty shopfront.
Neither Washington nor Brussels touched that kind of problem this time. Only one of these stories touches on “AI getting things wrong”: the case mentioned above in which The Seattle Times and Newsday accuse AI of making up content and saying they wrote it. Take The Seattle Times: a newspaper with its own legal team found AI putting invented content under its name, and the route open to it was a federal lawsuit. And lawsuits are slow: The New York Times, which is also suing OpenAI, filed in December 2023, and nearly three years on there’s still no judgment.
Lawsuits run on a clock measured in years. AI answers go out every day. Someone asking an AI like ChatGPT, Gemini or Perplexity today who to go to in your field, or whether your product is any good, gets an answer today, built from whatever the AI can find. A newspaper with a legal team takes this kind of thing to court, but most companies won’t fund a cross-border lawsuit, and couldn’t afford to wait that long anyway.
What to do about it
Act, but focus on what AI says about you in its answers rather than on the lawsuits. You can start today: ask AI how it describes your business and whether it gets anything wrong, without waiting for any ruling.
AI describes a business by checking what your own website says against what press coverage and reviews say. Where the accounts agree, it goes with them; where they clash, it may answer from the wrong one. Take the bakery that moved. Its website switched to the new address long ago, but as long as that old blog post stays online, AI may still repeat it.
When AI gets you wrong, nobody tells you. The Seattle Times at least found out for itself. Most businesses run into a different kind of error: AI treating old information, or someone else’s mistaken account, as you. Customers who read it rarely come back to mention it, and the AI sends no notice. So to learn which lines about you are wrong, you have to go and ask.
Once you find a wrong line, the hardest part is tracing it back. AI rarely says where it picked something up. Ask in different words, or ask a different AI, and the same wrong line turns up again. And even if you manage to get one mistaken article corrected, the line doesn’t necessarily go away. The later you catch it, the more customers have already read it.
Book a demo: set up a session with one of our consultants and we’ll run a round of the questions customers in your field actually ask through AI, live. Wherever a line is wrong, you’ll spot it at a glance.
Further reading
- OpenAI said it couldn’t search its own training data, then deleted the logs — an earlier chapter of the same case
- A court says robots.txt isn’t a lock — the technical way of blocking AI doesn’t count for much in court either
- Do you have a say in how AI uses your content? — where licensing and paid access actually stand