This week saw big changes in how people trust AI helpers. OpenAI launched a new ChatGPT Agent that can do tasks like shopping online for you. But they warned it's experimental and might make mistakes, so they added safety features like blocking dangerous requests.

A study by Capgemini found trust in AI agents is dropping fast worldwide. Only 27% of people trust them now, down from 43% last year. In the UK, trust fell from 50% to 32%. Most companies haven't fully used AI agents yet - only 2% have them working everywhere.

Amazon announced new tools called AgentCore to help businesses build safer AI helpers. They're spending $100 million to improve this technology. Experts say humans will still be needed to make big decisions and check the AI's work.