Topic starter
10/07/2026 9:24 pm
TL;DR: RSAC 2026 exposed a familiar problem in new clothing: security AI can only act safely if it reasons from complete, relationship-aware context, not just telemetry and logs, according to JupiterOne. Machine-speed remediation without the event history, identity links, and control exceptions behind a finding turns automation into a faster way to make the wrong decision.
Editorial analysis by NHI Mgmt Group, based on content published by JupiterOne: “AI is Everywhere at RSAC. Accuracy is Not.”.
Key questions
Q: What breaks when agentic AI is deployed without formal security policies?
A: Without formal policies, agentic AI can accumulate access, generate defects, and trigger unsafe actions faster than security teams can govern them.
Q: Why do incomplete identity and asset relationships increase agentic AI risk?
A: Because the agent needs to know who or what owns access, what that access reaches, and whether the permission still serves an approved purpose.
Q: How can security teams tell whether AI remediation is safe to automate?
A: They should ask whether the system can resolve ownership, dependency impact, control exceptions, and recent change history before it acts.
Practitioner guidance
- Map decision-critical context before enabling agentic response Identify the identity, asset, control, exception, and dependency data an agent must see before it can remediate a finding.
- Require ownership and lifecycle signals for every privileged identity Tie accounts, roles, service identities, and workload permissions to an owner, purpose, and expiry condition so the agent can distinguish active access from stale or intentionally temporary access.
- Block autonomous remediation when compensating controls exist Mark exceptions, board-approved risk acceptances, and downstream dependencies as hard constraints in the response workflow so the agent does not correct a local issue by breaking a wider control objective.
Bottom line: Agentic security fails when systems can only see current-state telemetry and not the relationships that explain why the state exists.
Explore further
View Full Forum → | NHI Foundation Course → | Our Services → | Read the full analysis →
10/07/2026 9:46 pm
Agentic AI security now has a context problem, not a model problem. The industry is demonstrating AI that can prioritise and respond, but those actions still depend on whether the system understands the environment it is changing. Security AI without relationship context will keep confusing state with meaning, which is where automation failures start. Practitioners should evaluate agentic systems by the completeness of the decision context they can consume, not by the fluency of their outputs.
A question worth separating out:
Q: Why do knowledge graphs matter more than dashboards for agentic security?
A: Dashboards summarise state, but knowledge graphs model relationships. That matters because security decisions depend on how assets, identities, controls, and dependencies connect, not just on whether a signal exists. A graph lets an agent trace blast radius and exception status before it acts, which is what makes autonomous response governable instead of just fast.
👉 Read our full editorial: Agentic AI security depends on context, not just telemetry