How AI agents are wrecking havoc in legacy security setups and enterprises are catching up (www.techradar.com)

A recent report from Mimecast at RSAC 2026 reveals a significant security gap within Fortune 500 companies, where 80% have deployed AI agents but only 14% have achieved full security approval. These AI agents operate autonomously with persistent credentials and access sensitive data without adhering to a security framework designed for human users. Traditional role-based security systems are proving inadequate as these agents can accumulate permissions and operate continuously, creating vulnerabilities that companies are struggling to manage. The implications of this gap are severe, with 97% of organizations suffering from AI-related breaches lacking proper access controls and 63% having no governance policies in place. As the speed of cyberattacks increases, as noted by CrowdStrike's findings of a record 27-second adversary breakout, enterprises must adapt their security measures to evaluate access requests in real-time. This involves integrating context-aware security that links each request to the user's entitlements and the task at hand, providing a dynamic enforcement layer that reacts to threats more rapidly than traditional approval processes. The focus must shift from slowing down AI agents to implementing smarter security measures that can accurately assess risk and manage access effectively.