RAD Security joins NVIDIA Inception for agentic security operations

At a glance

What this is: RAD Security’s announcement says its agentic RADBots use cloud and runtime telemetry to turn security signals into triage and remediation steps, while NVIDIA Inception provides the startup programme context.

Why it matters: For IAM and security teams, this matters because agentic security workflows blur the line between automation and authority, forcing programme owners to decide when a digital worker needs lifecycle and access governance.

Context

RAD Security’s announcement is about agentic security operations, where software systems investigate signals, prioritise findings, and move toward remediation using live telemetry. The identity question is not whether the workflow is automated, but whether the system is being allowed to act with operational authority.

That distinction matters because once a digital worker can suggest and then perform next steps, it begins to resemble a governed non-human identity rather than a passive analytics tool. The article frames RADBots as real-time security workers, which makes lifecycle, oversight, and decision boundary design part of the governance discussion.

NVIDIA Inception is the business context, but the security issue is broader: how organisations classify, monitor, and constrain agentic systems that sit between detection and response. This is a governance pattern that will look familiar to IAM, PAM, and NHI teams before it becomes a tooling question.

Key questions

Q: How should security teams implement agentic security without losing control?

A: Start in observe mode, limit the agent to investigation support, and gate any response action behind human approval. Define explicit boundaries for identity, endpoint, cloud, and SIEM access, then test evidence quality, false-positive impact, and analyst override behaviour before expanding scope.

Q: Why do agentic security workflows require lifecycle governance?

A: Because their authority can change as workflows, models, and integrations change. If a digital worker can investigate, prioritise, and act, then ownership, scope review, and retirement are lifecycle problems, not just deployment tasks. Governance has to follow the worker’s operational authority across its full active life.

Q: What are the signs that a security agent has too much authority?

A: The warning signs are direct action paths, unclear approval gates, and the ability to alter security state from telemetry alone. If the agent can move from detection to remediation without an explicit human checkpoint for high-risk steps, authority has outgrown observability.

Q: When should teams keep agentic security systems advisory only?

A: Keep them advisory when the cost of a wrong action is higher than the cost of a slower decision, especially in containment, access changes, and incident closure. Advisory mode preserves human accountability while still letting the system accelerate triage and recommendation.

Technical breakdown

Where agentic security operations cross the identity boundary

Agentic security operations sit between analytics and execution. A traditional automation pipeline follows predefined rules, but an agentic workflow can evaluate telemetry, choose next steps, and act without waiting for a human to click through every stage. That creates an identity problem because the system is no longer just processing data. It is making decisions that affect access, remediation, and containment. Once that happens, the questions change from performance and detection quality to authorisation scope, auditability, and delegation. If the agent can recommend and then perform actions, the control boundary must describe what it may do, when, and under whose accountability.

Practical implication: Treat action-taking security agents as governed identities with explicit authorisation boundaries, not as ordinary automation jobs.

Why deep telemetry changes the trust model

Deep cloud and runtime telemetry can improve the quality of agent decisions because the system is acting on fresher, more contextual evidence. But telemetry quality does not eliminate governance risk. If the agent can chain evidence gathering, triage, and remediation suggestion into a single runtime loop, then the trust question becomes whether the output is merely advisory or whether it can trigger downstream action. The more a system relies on real-time operational data, the more pressure there is to define event boundaries, approval gates, and rollback paths. In practice, telemetry-rich agents need controls that distinguish observation from execution.

Practical implication: Separate observation rights from execution rights so telemetry access does not become a proxy for remediation authority.

What digital workers mean for lifecycle governance

When a system is described as a digital worker, the governance model should move closer to identity lifecycle thinking. That means defining provisioning, scope, review, and retirement for the agent itself, not just for the account it uses. The risk is not only privilege excess. It is also unclear ownership when the agent’s behaviour evolves through workflow tuning, model updates, or new operational playbooks. For identity teams, this is where AI agent governance intersects with PAM and NHI lifecycle management: if the system can make decisions that affect security state, its authority must be inventoried and periodically revalidated.

Practical implication: Inventory agentic workers by business function and review their decision scope as part of lifecycle governance.

NHI Mgmt Group analysis

Agentic security operations create an identity boundary problem, not just an automation problem. The article describes RADBots as systems that investigate, prioritise, and act on security signals, which means they are moving from analytics into delegated operational authority. That is a different governance category from scripted SOAR-style automation because the workflow includes runtime decision-making. For practitioners, the practical issue is where to draw the line between a tool that informs response and a digital worker that participates in response.

Runtime telemetry becomes an authorisation input when the agent can act on it. Cloud and runtime data improve decision quality, but they also increase the number of moments where the system can justify action without human review. That changes the control objective from simply logging agent behaviour to governing which telemetry-fed decisions can become security actions. The field needs to stop treating decision quality and decision authority as the same thing.

Digital workers should be managed as lifecycle-bearing identities. Once a system can suggest remediation and perform next steps, its access, scope, ownership, and retirement require explicit governance. This is where NHI practice extends naturally into agentic AI governance, because the issue is not model intelligence alone but accountable authority over time. Security programmes that cannot inventory the agent cannot govern the agent.

Agentic security workflows will force IAM and PAM teams to redefine delegated authority. The article’s framing points to a market where more security tasks will be handed to systems that are neither static service accounts nor human analysts. That collapses older assumptions about who acts, when they act, and how review happens. The practitioner conclusion is clear: if the system can initiate remediation, it needs a governing model that matches that authority.

Agentic security operations need auditable limits before they need broader autonomy. The most important design question is not how much the system can do, but which actions are explicitly bounded, reversible, and attributable. In practice, that means operational teams should treat these workers as controlled actors in the security operating model, with defined escalation paths and explicit human oversight for higher-risk steps.

What this signals

Agentic security tools force a new control question: is the system merely assisting response, or is it already participating in decision execution? That distinction will shape how IAM, PAM, and NHI teams classify authority, review scope, and approve remediation.

Runtime telemetry is becoming a governance input, not just a detection input: once an agent can act on live cloud data, the programme must decide which evidence can justify action and which still requires human oversight. Security teams should expect this boundary to become a policy issue, not a product feature.

Agentic workflows will create pressure to manage digital workers the same way identity teams manage other non-human actors: by owner, scope, review cadence, and retirement path. The organisations that get ahead of this will be the ones that define authority before they expand autonomy.

For practitioners

• Define the agent’s authority boundary Document exactly which actions the security agent may suggest, initiate, or complete without human approval, and separate those rights from observation-only access.
• Inventory agentic workers as governed identities Create an inventory of digital workers, the accounts they use, the data they can access, and the business functions they support so ownership is explicit.
• Separate telemetry access from remediation authority Allow the agent to consume cloud and runtime telemetry without implicitly granting permissions to change security state, revoke access, or trigger containment.
• Add review gates for high-risk actions Require human review for actions that alter access, delete resources, or change incident status, even if lower-risk triage steps remain automated.
• Revalidate workflows after model or playbook changes Treat tuning, model updates, and workflow expansion as lifecycle events that can widen an agent’s effective scope and require re-approval.

Key takeaways

• Agentic security operations move beyond scripted automation when a system can investigate, prioritise, and perform next steps on its own.
• The core governance issue is authority, not telemetry volume, because live data becomes more sensitive once it can trigger action.
• Practitioners should inventory digital workers, scope their permissions, and require approval boundaries for any step that changes security state.

Key terms

• Agentic SecOps: A security operations model in which AI systems can coordinate tasks and take bounded actions across alert handling, investigation, and response. The critical question is whether those actions are governed, explainable, and reversible enough to fit enterprise control requirements.
• Digital Worker Identity: A non-human identity used by an automated system, AI agent, or other software-driven worker that can access tools or data. For MSPs, the key issue is that it must be governed like any other identity, with scoped access, monitoring, and revocation.
• Delegation authority: Delegation authority is the right for one identity to act on behalf of another within a defined scope. For agents, it must be explicit, time-bound, and auditable because the system may initiate actions independently once granted access.
• Runtime telemetry: Observation of what a system actually does while it is executing. In agentic CI/CD, this means seeing which commands, files, tools, and credentials an agent touched so security teams can detect misuse that static workflow review will miss.

Deepen your knowledge

NHI governance, agentic AI identity, and machine identity security are core topics in our NHI Foundation Level course, the industry's only accredited NHI security programme. If you are building or maturing an IAM programme, it is worth exploring.