OWASP: Agentic AI is already a security issue, no longer a future scenario
Agentic AI is developing faster than organizations can manage; the primary risks have now been proven in real incidents and CVEs. Prompt injection remains the most significant attack vector and forms the basis for many attacks on AI agents.
The line between AI Safety and AI Security is blurring: the same design choices cause both errors and security incidents. Agentic supply chains have become a new attack surface, involving the exploitation of MCP servers, skill registries, and AI frameworks.
Coding agents dominate the market and are gaining increasing autonomy over repositories, CI/CD pipelines, and cloud environments. Shadow AI is growing explosively: approximately half of all employees use AI tools outside the visibility of IT and security.
Non-Human Identities (NHIs) and AI agents are creating a new identity crisis as they are granted increasing rights and access. Governance is shifting from periodic audits to real-time monitoring, automatic detection, and immediate kill switches.