What 2025 taught us about AI agent security: A practitioner’s guide to the incidents shaping enterprise adoption

This paper examines security incidents affecting artificial intelligence (AI) coding assistants and enterprise AI agents during 2025, providing security teams with practical guidance for risk assessment and mitigation. The paper synthesises notable vulnerabilities and public advisories affecting major platforms including GitHub Copilot, Cursor, Amazon Q Developer, Microsoft 365 Copilot and Claude Code, with a focus on enterprise-relevant impact rather than exhaustive coverage. Three vulnerability patterns recur across platforms: prompt injection enabling privilege escalation, inadequate authentication at trust boundaries, and insufficient isolation between AI operations and sensitive resources. These patterns enabled remote code execution, credential theft and data exfiltration. Organisations deploying AI agents require updated security controls including agent inventory management, configuration hardening and vendor security assessment. The recurring patterns across diverse tools indicate the need for new forms of trust infrastructure and governance beyond product-by-product patching. This article is also included in The Business & Management Collection which can be accessed at https://hstalks.com/business/.