Courts, states and data rules close in on AI companies

It is no longer just engineers who decide how AI should behave. This week it was politicians, prosecutors, rights holders and even the pope.

One clear thread is the pressure on OpenAI. The Texas attorney general has sued the company and CEO Sam Altman, arguing that they misled the public about ChatGPT's safety. Among other things, the state demands that ChatGPT not be offered to minors in Texas. According to Spotlight Delaware, Delaware has also sent an investigative demand about safety and governance, after California had already done something similar. Meanwhile, OpenAI plans to use data from Daybreak Blue sessions for training starting November 29, which has researchers worried that their safety work will be used to improve the models. Following criticism, a separate opt-out has been promised.

Who gets to learn from what?

Another thread is about copyright. The Elvis Presley estate and other rights holders have sued Anthropic, claiming that Claude was trained on 274 songs without permission. Wikimedia is proposing a middle ground to a US court: Training should probably count as "fair use," meaning permitted use without a separate agreement, while verbatim copying in chatbot answers should not. Anthropic has itself given Claude Code a rule meant to prevent copies of someone else's specific game, but it is unclear how well it works.

The industry is also trying to solve some of this on its own. Google, Anthropic, OpenAI and Microsoft plan to back a new signal in robots.txt, making it easier for sites to prohibit AI training. However, it only works going forward.

Power, neutrality and limits

A third thread is about whom AI answers serve. When models are forced to pick a candidate in US elections, Grok most often chooses the Republican, while GPT almost always chooses the Democrat. The consumer versions, however, refused to answer. One study also found that the chatbots often linked to anti-abortion sites without saying who was behind them, and 68 members of the European Parliament have questioned the Commission.

Then there is security and control. Sweden is blocking ChatGPT and Claude on government work computers and pointing to Copilot instead. In the US, two Democrats are demanding that the White House preserve the president's conversations with chatbots, following a report on the use of Grok. China has proposed requiring prior approval before advanced models are shared abroad, but it has not been adopted as a requirement, while the US already bars DeepSeek from civilian government procurement. The pope is calling for a ban on lethal autonomous weapons.

Rules are also taking effect in small ways: Starting November 7, Grok's API will add an invisible watermark to text in order to comply with the EU's AI Act.

What can you do?

Check whether your workplace has rules on which chatbots may be used and what you may write to them. Read the data-sharing settings in the tools you use, and look for opt-outs. And remember that a chatbot's answer is not neutral just because it sounds confident. Keep an eye on the Texas case, the Anthropic lawsuit and the first concrete opt-out deadlines.