Topic starter
19/08/2026 3:38 pm
TL;DR: Agentic AI security now spans identity, memory, tool access, and runtime enforcement, because agents can plan, call APIs, write data, and persist instructions across workflows without constant oversight, according to Cycode. The governance problem is that access review models assume stable, reviewable behaviour, but autonomous action can create privilege use and leakage before any human sees it.
Editorial analysis by NHI Mgmt Group, based on content published by Cycode: “What Is Agentic AI Security?”.
By the numbers:
- 88% of organizations experienced confirmed or suspected AI agent security incidents in the last year.
- Only 22% of teams treat agents as independent identities, breaking audit trails and accountability.
Key questions
Q: How should security teams govern autonomous agents that use backend tools?
A: Treat autonomous agents as identities with their own scopes, policies, and revocation paths.
Q: Why do AI agents complicate traditional access reviews?
A: AI agents complicate access reviews because they can accumulate permissions across tools and environments faster than manual certification cycles can observe.
Q: What breaks when agent memory is shared across users or workflows?
A: Shared memory turns prior interactions into a hidden input to later decisions, which means one poisoned or sensitive context can influence other sessions.
Practitioner guidance
- Define agent-specific identities Issue distinct credentials per agent and stop treating agents as extensions of human users or generic service accounts.
- Constrain tool execution paths Inventory every approved tool, API, and MCP server an agent can invoke, then remove anything that is not required for the task.
- Isolate memory and context stores Separate memory by user, session, and workflow so one interaction cannot seed another.
Bottom line: Agentic AI security is fundamentally about governing autonomous action, not just filtering model output.
Explore further
View Full Forum → | NHI Foundation Course → | Our Services → | Read the full analysis →
19/08/2026 4:18 pm
Agentic AI security is now an identity governance problem, not just an AI safety problem. The article shows that agents can plan, call tools, and persist memory across workflows, which means the security question is who or what is authorised to act. That pushes governance into identity, privilege, and lifecycle territory that traditional model-centric controls do not address. Practitioners should treat agent behaviour as an access-control issue first.
A few things that frame the scale:
- 53% of security leaders expect AI to run major portions of their infrastructure autonomously within the next three years, according to the 2026 Infrastructure Identity Survey.
A question worth separating out:
Q: What is the difference between prompt-level guardrails and runtime guardrails for AI agents?
A: Prompt-level guardrails inspect what the user asks or what the model says. Runtime guardrails inspect the whole interaction while it is happening, including tool requests, tool responses, and downstream actions. That matters because many agentic attacks are chained and only become visible when the full execution path is evaluated as one session.
👉 Read our full editorial: Agentic AI security exposes identity, tool and memory risks