AI Agent Security Threats Escalate: Hackers Exploit, Microsoft Adds Controls

Hackers reportedly used AI agents to breach 100-plus company sites with minimal involvement. Microsoft is reportedly rolling out enterprise controls to risk-score, approve or shut down agents in Microsoft 365. A post citing research says asking models to repeat attacker-written tool calls got six major agent frameworks to run them; it says the same paper found 253 bugs across 118 MCP servers.