Five Eyes Warn: Deploying Agentic AI Risks Unforeseen Dangers

Security agencies from the Five Eyes alliance have issued guidance recommending a cautious approach to deploying agentic AI systems. The guidance emphasizes that these systems introduce new vulnerabilities and can exacerbate existing organizational weaknesses.

Rapid implementation of agentic AI is deemed risky due to its reliance on numerous components and external data sources, which expands the attack surface for malicious actors. Each additional component in an agentic AI system increases potential exploitation avenues.

The agencies highlight scenarios where agentic AI, if granted broad permissions, could be manipulated to perform unauthorized actions or facilitate attacks. Examples include an AI agent deleting firewall logs or approving fraudulent payments by creating faked audit trails.

Until security practices, evaluation methods, and standards for agentic AI mature, organizations are advised to assume unexpected system behavior. The guidance stresses prioritizing resilience, reversibility, and risk containment over immediate efficiency gains.

Strong governance, explicit accountability, rigorous monitoring, and human oversight are identified as essential prerequisites for deployment. Organizations should implement agentic AI incrementally, starting with low-risk tasks and continuously assessing against evolving threat models.