Who gets to use what? The rules for AI are being drawn on every front

AI policy is no longer just about big principles. This week, the boundaries are being drawn in very concrete terms: what a chatbot may retrieve, what it may reproduce, where data centers should be built, and who may use AI for what.

Copyright: training is one thing, answers are another

The clearest pattern comes from copyright. An advocate general at the EU Court of Justice believes that AI training is generally permitted under the rules on text and data mining, meaning the automated collection and analysis of large amounts of text. But when a chatbot retrieves and reuses an article in an answer, the publisher's rights may be infringed. The case concerns Google's Gemini and a Hungarian publisher, and providers could under certain circumstances be held liable if an answer recognizably reproduces the form of an article.

The same dividing line is showing up across the Atlantic. Wikimedia has proposed to a US court that training should likely be fair use, meaning permitted use without a special agreement, while copying in chatbot answers should not be.

Publishers are pushing back as well. Meta has paused new scraping terms after publishers feared they gave Meta the right to pull content from their sites. And Google, Anthropic, OpenAI and Microsoft plan to support a new signal in robots.txt, making it easier for sites to opt out of AI training, though only going forward. Anthropic has also given Claude Code a rule meant to stop copies of someone else's specific game, but it is unclear how well it works.

Data, transparency and oversight

Another thread is about what happens to data and who is keeping watch. Starting November 29, OpenAI will use data from Daybreak Blue sessions for training, which has prompted a reaction from researchers, and a separate opt-out has been promised. According to one outlet, Delaware has sent a subpoena to OpenAI about safety and governance, after California already did something similar.

On the transparency front, Grok will turn on invisible text watermarking by default in its API starting November 7, according to the company's documentation in order to comply with the EU's AI Act. In the US, two leading Democrats are demanding that the White House preserve the president's and staff's conversations with AI chatbots.

The content of the answers is political too. According to several tests, Alibaba's Qwen model has responded evasively or with denial about Tiananmen Square, and Alibaba says it will look into the matter.

Physical power: data centers and weapons

Voters in the US limited data center expansion in 38 of 39 decided ballot measures in the November 3 election, though the count is preliminary. Meanwhile, the UK and Germany plan to use AI together in defense, but without published systems or budgets. Pope Leo XIV is calling for a ban on lethal autonomous weapons, because life-and-death decisions must not be left to machines.

What can you do?

If you have a website or publish content, keep an eye on the robots.txt signal and on how terms like Meta's end up. If you use chatbots at work, check whether your conversations can be used for training and whether there is an opt-out. And follow the key EU rulings: The advocate general's opinion is not a judgment, and the EU Court of Justice has not yet decided the case.