What AI Changes About the Cybersecurity Skills Your Team Needs

Every security task automated changes more than the time it takes to complete the work. It changes where accountability sits, how practitioners acquire experience, and where the organization remains dependent on human judgment.

That makes AI adoption a workforce-design decision. CISOs need to define how people and AI divide responsibility, which decisions require human authority, and how employees will develop expertise when automation removes work they once learned from.

The same technology is increasing adversaries’ ability to scale operations and imitate trusted behavior. Teams need the technical depth to understand that activity and the judgment to challenge a plausible answer, connect separate weaknesses, and recognize when a familiar-looking action serves a malicious objective.

A workforce strategy built around stable job descriptions and external hiring will struggle to address those changes. Leaders need to reconsider the work itself and deliberately develop the people who will remain accountable for it.

The scale of the development challenge is already visible. In the 2025 ISC2 Cybersecurity Workforce Study, 95% of cybersecurity professionals reported at least one skills gap on their team, and 41% identified AI as the most pressing gap. Headcount alone cannot show whether a team has the capabilities its environment requires.

Consider a workflow that compares external threat intelligence with internal asset inventories and vulnerability rankings. AI can assemble that information and produce a risk-based order of work much faster than a manual process.

The analyst’s responsibility moves toward evaluating the result and understanding if the inputs are sufficient, if the prioritization reflect business condition and is the recommended action justified.

Without explicit ownership of those decisions, automation can accelerate a weak process. A prioritized list may look authoritative while overlooking an incomplete inventory or a business dependency the model was never given.

As AI-generated code and analysis reduce handoffs across engineering, testing, and security, leaders need to define capabilities around required outcomes. Producing convincing output tells them less about someone’s readiness when a tool can generate it. They need evidence of how the employee framed the problem, tested the result, and decided what should happen next.

For each AI-assisted workflow, assign who frames the objective, who can challenge the output, who authorizes action, and who remains accountable when the result is wrong.

Entry-level work has historically given practitioners repeated exposure to bounded cases. Reviewing alerts, escalating unusual activity, and observing how experienced colleagues handled mistakes helped them learn which details mattered.

When AI absorbs more of that work, the organization also changes its succession model. It may gain efficiency while weakening the pipeline that produces future investigators, engineers, and security leaders.

Leaders should identify the judgment those tasks developed and recreate opportunities to practice it. Hands-on labs, challenges, guided reviews, and open-ended assessments can expose employees to consequential decisions in a safe environment.

Early exercises might isolate one decision, and later practice can introduce incomplete information, conflicting signals, or activity spanning several systems. Progression should depend on how employees perform as those demands increase.

The objective is to preserve the experience that creates expertise. Training hours and time in role cannot establish whether someone has developed it.

External hiring remains important when an organization lacks necessary expertise. It becomes an expensive dependency when every new security requirement triggers another search for a specialist.

PwC’s 2025 Global AI Jobs Barometer found that workers with AI skills commanded an average wage premium of 56% in 2024. Skills sought in AI-exposed roles were changing 66% faster than those in less-exposed roles. Even a successful search leaves leaders with a continuing development requirement.

Existing employees already understand the organization’s systems, operations, and priorities. Someone with experience in application security, cloud, identity, or offensive security may have foundations that can be extended into AI security responsibilities.

The workforce decision is which gaps genuinely require recruitment and which can be addressed through structured development. Leaders also need to know whether critical capability is concentrated in one or two employees. Filling a vacancy does little to reduce that dependency if the knowledge remains isolated.

Building broader internal coverage requires a transferable approach to security problems. Offensive thinking provides that foundation.

Security functions have distinct responsibilities. An attacker pursues an objective across whatever boundaries the environment allows.

A weakness in an application may provide initial access. Exposed credentials may extend that access. A trust relationship may then open a route into a sensitive system. Each issue can appear limited when reviewed alone.

Offensive thinking means examining how those weaknesses could connect, how defenses could be evaded, and how expected behavior could be manipulated. Practitioners across security roles need enough understanding of that progression to evaluate the part of the environment they protect.

Useful questions include:

• What can be reached from the initial point of access?
• Which assumptions about trust or permissions can be challenged?
• How could separate weaknesses be connected?
• What would allow the attacker to remain undetected?
• Which path could create the greatest impact?
• How could malicious activity be made to appear normal?

These questions test assumptions routine reviews may leave unchallenged.

The presence of a control and the ability to defend an environment are different things. Teams need practice examining the complete attack path to understand how a weakness in one area affects another.

A CISA red team assessment of a federal civilian executive branch organization illustrates the gap. The team compromised the Windows domain and accessed sensitive systems. Five months into the assessment, CISA notified the security operations center because the activity remained undetected. The assessment exposed weaknesses in segmentation, logging, detection, and coordination among decentralized teams.

Testing the connected environment revealed defensive gaps that the presence of individual controls had not resolved.

The same judgment helps leaders decide where AI autonomy should stop.

Isolating a device after malware has been conclusively identified may fit a defined automation policy: the trigger is specific, the response is contained, and the action can be reversed. That decision changes when the evidence is ambiguous or the response could interrupt a critical service.

Leaders need practitioners who can recognize those differences and intervene. That requires technical understanding of the system and enough business context to judge the consequences of an action.

An attacker using valid credentials and familiar access patterns may produce few obviously suspicious signals. Recognizing the threat can require evidence from several teams and an understanding of the objective connecting the activity.

Offensive thinking gives those teams a shared way to examine the problem.

Developing that judgment requires contact with the work. Practitioners need to select tools, follow evidence, revise failed approaches, and defend their conclusions. 

Those foundations also matter for AI red teaming. Practitioners still need to understand the applications, APIs, identities, infrastructure, and data surrounding an AI system to determine how an AI-specific weakness could contribute to a wider compromise.

The next step is to turn those workforce decisions into a development plan: define the responsibilities employees will own, create the experience they need, and establish how they will demonstrate readiness.

In next article, Redesign the Workforce Before AI Does It for You, we introduce the AI-Ready Cyber Workforce Model and show how to apply it, using AI red teaming as a practical starting point.