A major new report on artificial intelligence safety came out this week, and it has important information about AI agents - computer programs that can think and make decisions on their own. The report says that AI agents are becoming more powerful, but they also create new safety challenges.

AI agents can now do some complex tasks like writing code and planning projects, but they still need help from humans. Scientists worry that as these agents become smarter, it will be harder for people to stop them if something goes wrong. One big concern is that agents might make mistakes before anyone can fix them.

The safety report also shows that people at big AI companies are worried about keeping users safe. Several important safety workers left their jobs this week because they thought their companies were moving too fast and not being careful enough. One researcher at OpenAI said the company is creating new products like ads without making sure they are completely safe.

Companies like OpenAI and Anthropic have released many new AI tools this year, including agents that can help with writing code and doing daily tasks. But experts say these companies need to focus more on safety and less on making new products so quickly. The new safety report will help world leaders talk about how to keep AI agents safe as they become more powerful.