This week saw big steps in keeping AI safe and fair. In Europe, new rules called the EU AI Act started with fines against companies using unfair hiring AI. The European Commission also shared guidelines to help companies follow these rules before they fully start in August.

A new Global AI Ethics Observatory launched by UNESCO and Anthropic now tracks AI problems like deepfakes worldwide. This helps spot unfair AI faster.

Experts warned about security risks with AI agents that work on their own. Problems include:

• Shadow AI where hidden agents access private data
• Overprivileged agents having too much access
• Prompt injection attacks tricking AI into bad actions

Companies like Deloitte and HCLTech shared safety plans for AI agents. They suggest three guardrails: basic rules, extra rules for risky AI, and society-wide protections.

Meta faced privacy worries over its brain-reading headset that controls AI. People want open rules for such powerful tech.

A report found only 2% of companies fully use agentic AI safely, even though it could create $450 billion by 2028. Building trust remains key for AI's future.