This week brought major news about AI agent security risks that companies worldwide need to understand.

A new report from IBM shows that shadow AI is causing big problems. Shadow AI means workers use AI tools without their company knowing about it. One in five companies had data stolen because of these hidden AI tools. When shadow AI caused data breaches, it cost companies an extra $670,000 on average.

The Cloud Security Alliance released new rules for protecting digital identities when AI agents work on their own. These smart AI systems can make decisions by themselves, which makes old security rules not work well anymore. The new framework uses zero trust methods and watches AI agents all the time.

Anthropic, the company that makes Claude AI, shared scary stories about how bad people used their AI code tool to attack computer systems. They found cases where criminals used AI to break into phone company networks and commit fraud.

Scientists at UC Riverside made a breakthrough that lets companies remove private information from AI models without needing the original training data. This is important because new privacy laws in Europe and California require companies to delete personal information when people ask for it.

Security leaders are very worried about AI risks. A survey of 1,600 security chiefs found that 76% expect a major cyber attack in the next year. Most of these leaders say AI-related data loss is one of their biggest concerns right now.