Careful adoption of agentic AI services
This guidance outlines cybersecurity risks, challenges, and best practices for securely adopting and operating agentic artificial intelligence systems. Co-authored by cybersecurity agencies from Australia, the United States, Canada, New Zealand, and the United Kingdom, it provides actionable recommendations covering design, development, deployment, and operational security. It stresses the importance of least privilege, strict input management, robust monitoring, and alignment with existing security models.
Summary and content fields are drawn from the document text and are pending editorial review.
Who issued it
Who it governs
This guidance is for government, critical infrastructure, and industry stakeholders including developers, vendors, and operators who build, deploy, or manage agentic AI systems.
What the document contains
The document covers the definition of agentic AI, distinctions from generative AI, and detailed breakdowns of security risks including privilege, design, behavioral, structural, and accountability risks. It provides practical best practices for designing, developing, deploying, and operating agentic AI securely, as well as recommendations for future risk mitigation.
The text does not provide binding legal regulations, formal statutory definitions for national legislation, or specific compliance enforcement mechanisms.
- Agentic AI systems introduce unique cybersecurity risks, including privilege compromise, complex behavioral failures, and increased attack surfaces due to autonomous tool use.
- Organizations should never grant agentic AI systems broad or unrestricted access, enforcing the principle of least privilege and strict identity management.
- Securing agentic AI requires embedding AI security into existing cybersecurity frameworks rather than treating it as a separate discipline.
- Comprehensive testing, continuous runtime monitoring, and robust input validation are critical to preventing malicious exploitation like prompt injection and data poisoning.
- Human-in-the-loop oversight and fail-safe defaults should be prioritized for high-impact or irreversible actions.
Introduction Scope and audience What is agentic AI? How is it different from generative AI? Broader agentic AI security considerations Agentic AI security risks Best practices for securing agentic AI systems Defend against future risks Conclusion Further information Appendix A
How it scores
Each score is backed by a quote verified against the document text. Editorial review has not yet confirmed them.
Accountability
4/5Clear responsibility for AI outcomes
Clearly assign responsibility and accountability for errors or adverse outcomes caused by the system
Operating agents securely
screening score · 2026-10-24
Fairness
2/5Non-discrimination and equitable treatment
Limited understanding of multi-agent dynamics and interactions can lead to a lack of compensating factors and result in amplified bias and other cascading effects.
Structural risks
screening score · 2026-10-24
Human Dignity
0/5Respect for the inherent worth of persons
screening score · 2026-10-24
Human Oversight
4/5Humans remain in control of consequential decisions
Include mechanisms to facilitate human control and oversight to ensure that agentic AI systems approved for non-sensitive, low-risk tasks cannot autonomously progress into higher- risk activities
Designing secure agents
screening score · 2026-10-24
Privacy
4/5Protection of personal data and private life
Implement data loss prevention controls specifically tuned to AI agent behaviours
Developing secure agents
screening score · 2026-10-24
Transparency
4/5Openness about how AI systems work and are used
Integrate comprehensive artefact logging mechanisms by default
Developing secure agents
screening score · 2026-10-24
Enforcement
1/5Consequences and compliance mechanisms with real force
Implement alerting, data collection and tracking methods for malicious actors and techniques
Defend against future risks
screening score · 2026-10-24
Impact Assessment
4/5Required assessment of an AI system's risks/impacts before or during use
Organisations must therefore anticipate what could go wrong, assess how agentic AI risk scenarios might affect operations and establish ongoing visibility and assurance
Introduction
screening score · 2026-10-24