• About Us
  • Contact

From AI-assisted to agentic: The next stage of security operations

Published: 19th November 2026

Can your SOC respond at the speed of an
AI-enabled attack?
 

In a recent episode of Red Helix’s Cyber in Focus podcast, Tom Exelby spoke to James Jenkins at CrowdStrike about the emergence of the agentic SOC and what it could mean for security operations. Their discussion highlighted a fundamental shift: AI is no longer simply a tool to help analysts complete individual tasks. It is beginning to change how investigations are conducted and decisions are supported. 

For IT and security leaders, the question is not simply whether to introduce AI into the SOC. It is how to use agentic capabilities to improve response, manage risk and give security teams greater capacity without surrendering control. 

Understanding the difference between
AI-assisted and agentic SOCs
 

One of the central points in the discussion was the need to distinguish between four models: the traditional SOC, the AI-assisted SOC, the agentic SOC and the autonomous SOC. 

In a traditional SOC, analysts work through alerts and investigations, often across tiered levels of responsibility. Level one analysts typically triage alerts, with more complex investigations escalated to experienced colleagues. 

An AI-assisted SOC introduces tools that help analysts carry out tasks such as writing queries, summarising incidents and interpreting information. These capabilities can reduce manual work, but the analyst remains responsible for directing each step. For example, asking an AI model to write a query, does not make the SOC agentic if the analyst still has to drive the investigation. 

The distinction is the ability to give an agent a goal and allow it to determine the next investigative steps. An agent can reason through a task, use tools within the SOC environment and work towards an outcome within defined boundaries. 

That shift from answering questions to conducting investigations is what makes the agentic model different. 

The fourth model, a fully autonomous SOC, would remove humans from the decision-making process, this is not a realistic option at present. Security teams remain accountable for decisions and outcomes, and human context is still required when assessing activity and business risk. 

For organisations evaluating security platforms, understanding these distinctions matters. A product described as agentic may simply provide
AI-assisted features. 
 

Trust requires visibility and control 

Giving AI agents the ability to investigate and act within security systems introduces a new governance challenge. 

Agents must be subject to the same core security principles applied to human users. They should operate with least-privilege access, and their activity must be auditable. 

That requires visibility into the prompts they receive, the tools they call and the actions they take. Security teams must be able to establish whether an agent is operating within its authorised boundaries and whether instructions from users or other agents are legitimate. 

This is not simply a matter of preventing an agent from making an incorrect decision. AI systems can also become part of the attack surface. If an attacker gains access to an organisation’s AI tools, they may be able to use the permissions already granted to those tools to retrieve sensitive information or carry out actions. 

The same capabilities that make agents useful to defenders can therefore create risk if access and oversight are poorly managed. 

The principal of “trust but verify” remains a practical basis for adoption, agents should be given defined responsibilities, restricted permissions and clear monitoring, with human oversight proportionate to the risk of the task. 

Context is what makes agents useful 

An agent’s ability to investigate depends on the quality of the context available to it. 

An analyst does not assess every alert in isolation, they build an understanding of the organisation, its users, normal activity and previous incidents. Agents require similar context if they are to reach useful investigative conclusions. 

For example, repeated alerts about unusual travel activity may be legitimate for an employee who regularly moves between locations for work. Without historical context, an agent may treat each alert as a new incident. With access to relevant previous cases and information about the user’s environment, it can make a more informed assessment. 

For security leaders, this points to a broader requirement. Introducing agents is not just a technology deployment. It involves making relevant operational knowledge available, maintaining the quality of security data and ensuring that investigations can be reviewed as circumstances change. 

Make the transition a business decision 

For IT and security leaders, the strategic decision is not whether to adopt AI for its own sake. It is how to build a security operation that can respond to changing threats at the required speed, while keeping accountability and control firmly with the people responsible for protecting the business.

Watch the full episode here

From AI-assisted to agentic: The next stage of security operations

Find out more

AI resilience: Turning AI adoption into a managed business risk

Find out more

What the Cyber Security and Resilience Bill could change when you use a managed security provider

Find out more

Thinking of switching your managed security provider?

Image of the Red Helix team during a meeting about managed cyber security
Find out more

Navigating Risk, Driving Resilience

Find out more

Learn How to Improve Your Law Firm’s Cyber Security

Lawyers using laptops in a bright office
Find out more

Cyber Connect

Find out more

The 47% cyber security confidence gap: responsibility is not the same as readiness

Find out more

Is Your MSSP Still Meeting Your Security Needs?

Find out more