Microsoft made big upgrades to its cybersecurity tools this week by adding 11 new AI agents to its Security Copilot platform. One agent specializes in spotting fake emails, checking 30 billion alerts each year, while another fixes security holes by processing 84 trillion daily signals. These agents work as a team, sharing information to block threats faster than human teams could alone.

Apple showed progress on Project Mulberry, a health coach agent that reads data from Apple Watches. It creates custom workout and diet plans, alerting users if their heart rate or sleep patterns seem off. Set to launch with iOS 19.4, this agent will mark Apple's first major push into AI-driven healthcare.

Ethics took center stage at the Boston Generative AI Meetup, where developers demonstrated agents that explain their decisions in simple terms. For example, one agent pair split tasks fairly between workers while keeping managers in the loop. These systems use real-time reasoning to balance efficiency with human oversight.

Business tools saw updates too. Google's Agent Garden now offers pre-built agents for supply chain tracking, while Amazon Bedrock added features for secure data sharing between agent teams. These platforms help companies deploy multi-agent systems without needing deep AI expertise.

In finance, autonomous investment agents from Microsoft's hackathon can rebalance portfolios instantly when markets shift. Another project created fraud-detection agents that cross-check transactions across banks, reducing false alarms by 40% compared to older systems.

As these systems grow, experts stress the need for clear rules. The Boston panel proposed mandatory ethics checklists for agents handling sensitive tasks like hiring or medical advice, ensuring humans always review critical decisions.