AI starts acting on our behalf, and now it calls for new rules

When many people talk about AI, they think of a chatbot that answers questions. This week was more about AI that does things on its own: fetching quotes and reading files on your computer, while Musk says Grok Bot will be able to call businesses. It is practical, but it raises a question about who actually decides what the AI does.

At the same time, the world is tightening its grip. One government is shutting out two well-known chatbots, one US state is going to court, and researchers are testing whether chatbots have political favorites. Here are the big themes of the week.

AI that acts on your behalf

An American insurer, Kin Insurance, has opened its website to AI agents, so paying ChatGPT users can get a quote for home insurance. An agent is an AI program that carries out tasks for you instead of just answering. The purchase still happens with Kin, and the agents cannot assess gaps in coverage American insurer lets ChatGPT fetch quotes on . That is an important caveat, because a cheap quote is not the same as good insurance.

Elon Musk says his Grok Bot will be able to call businesses, ask for something on your behalf and send you a summary. The feature is not documented, however, and pricing and rollout are unclear Musk: Grok Bot will be able to call businesses for you. It is wise to hold off on the excitement here.

Agents also have a downside. The security firm Forever Security warns of an attack in which a website places a file in the Downloads folder, which AI coding agents then follow. The success rates come from the company's own tests Hidden instruction file in Downloads can get AI coding agents to.

Rules, lawsuits and trust

In Sweden, the Government Offices will block ChatGPT and Claude on work computers starting Monday, October 26, 2026. The reason is information security, and employees are being directed to Microsoft's Copilot Swedish government shuts out ChatGPT and Claude on work comput. It is a clear signal that trust does not come on its own.

In Texas, the attorney general has sued OpenAI and CEO Sam Altman. The state believes the company misled the public about ChatGPT's safety, and demands, among other things, that ChatGPT not be offered to minors in Texas Texas wants ChatGPT away from children and demands safety checks. And the Elvis Presley estate has, according to sources, sued Anthropic, because Claude is alleged to have been trained on Presley's music without permission Elvis heirs take Anthropic to court over 274 songs.

Privacy is also at stake. Google's updated privacy policy takes effect on October 16, 2026, and according to one review, data from Gmail, Calendar and Photos can affect personalization if you connect them to Google Search Google's new privacy rules: Can your emails affect the ads.

When the chatbot has to pick a side

A study from Convovo shows what happens when an AI model is not allowed to abstain from picking a candidate. Grok picks Republicans most often and GPT picks Democrats most often Force a chatbot to pick a candidate, and it chooses a side. Researchers had the two models pick a candidate in nearly 500 American elections. Grok most often chose the Republican, while GPT almost always chose the Democrat Forced to choose, two AI models recommended opposite parties. The consumer versions refused to answer the question, though, so in everyday use you are unlikely to encounter this behavior directly. But it reminds us that the models are not neutral.

Cheap on paper, expensive in practice

Prices get a lot of attention, but comparisons can mislead. A widely shared comparison pits DeepSeek's $0.15 against Claude's $4 per million tokens, the small chunks of text that AI works with. The figures cover different models, points in time and kinds of usage, however Cheap AI on paper: Where the DeepSeek and Claude comparison falls short. If you want an overview of what AI coding help costs, prices at Claude, OpenAI and Google range from free to $500 a month AI coding help: What Claude, OpenAI and Google cost.

Big promises also need to be weighed. Elon Musk promised a full-length feature film made by AI in July, but Grok's version of the Odyssey has now been released as a first episode of just over seven minutes Musk shows seven minutes of AI Odyssey, but promised a full film.

AI on your own computer

If you worry about where your data ends up, local AI can be an answer. A copywriter got a small language model running on a 10-year-old mini PC with no graphics card. It worked, but the answers took a long time, and the smallest model made mistakes A 10-year-old mini PC can run AI, but it takes patience. Local AI is possible, then, but it takes patience.

What you can do yourself

When an AI offers to act for you, always check what you are getting, especially with insurance American insurer lets ChatGPT fetch quotes on . Ask at your workplace which AI tools are allowed, and why Swedish government shuts out ChatGPT and Claude on work comput. And take price comparisons with a grain of salt until you know what the figures actually cover Cheap AI on paper: Where the DeepSeek and Claude comparison falls short. Read the articles for the details.