AI Coding Agents Expose Critical Security Gap as Nearly Half of Enterprises Report Incidents

AI coding agents like Claude Code and GitHub Copilot are autonomously acting across enterprise systems, and nearly half of organizations have already experienced security incidents involving them—yet only 15% feel confident they can detect or respond, exposing a critical gap between what actions are permitted and what is actually appropriate.

• Coding agents like Claude Code, Cursor, GitHub Copilot, and Gemini CLI are autonomously browsing the web, writing to filesystems, committing code, and calling external APIs under engineer identities across enterprise environments right now.
• A critical security gap is emerging as existing tools like EDR, DLP, IAM, and SIEM are built to verify whether actions are permitted but are unable to determine whether those actions are appropriate, leaving enterprises exposed to attacks using legitimate credentials with malicious intent.
• 47% of security and IT professionals report experiencing a security incident involving an AI agent in the past 12 months, yet only 15% express high confidence in their ability to detect and respond to such incidents, highlighting an urgent need for intent-based observability and governance frameworks.