ProBackend
Cybersecurity
5 days ago6 min read

AI Cybersecurity Spending: Make the Bet Before the Proof?

AI Cybersecurity Spending: Make the Bet Before the Proof?

CISOs are prioritizing AI cybersecurity despite uncertain returns. Here’s how to understand its defensive uses and invest without mistaking urgency for evidence.

What is AI in cybersecurity?

AI in cybersecurity means using systems that identify patterns, classify signals, summarize information, or automate parts of security work. It isn’t one product category, and it doesn’t automatically make a security program smarter. The useful question is what job the system performs, what data it relies on, and where a person remains accountable.

In practice, AI can support security operations by helping analysts sort alerts and investigate threats; it can also be applied to identity and access management. Those are among the areas highlighted in the IANS and Artico Search spending coverage. In a defensive setting, the promise is less time spent on repetitive work and more attention available for judgment-heavy investigation. That is a plausible operational benefit, not proof that any particular tool prevents incidents.

The distinction matters. A security team may buy AI-enabled capability to deal with a changing threat environment, while still needing to test whether it improves its own outcomes. “AI” on a procurement slide is not a measurable control.

Why AI cybersecurity is attracting new money

The budget numbers make the decision hard to ignore. A survey of more than 500 CISOs found average security budgets grew 5% in 2026, after 4% in 2025, while the median budget was flat for a second consecutive year. Yet 69% of surveyed CISOs named AI their leading priority for new cybersecurity dollars. The same reporting says 24% gave AI a separate security budget line; 38% included it in broader security spending, and another 38% funded it from IT, data, or innovation budgets.

That’s a strong signal of attention, but not a clean verdict on value. When most budgets aren’t expanding much, a priority can mean reallocating scarce dollars rather than opening a bottomless new purse. And where AI is funded outside security, it may be harder to see its total cost or to assign responsibility for risk.

There’s a practical reason leaders may move before they have a tidy return-on-investment calculation: AI capabilities are already moving into production, and security leaders worry that attackers can use automation to operate faster. The Dark Reading reporting quotes practitioners describing AI-assisted phishing, malware mutation, and vulnerability hunting as concerns. These are arguments for readiness and defensive experimentation. They are not, by themselves, evidence that a rushed purchase will outperform a well-run existing control.

How AI is used in cybersecurity

Start with the work, not the label. The cited budget reporting points to security-operations automation and identity and access management as concrete areas of AI-related spend. In operations, a tool might help prioritize or investigate signals; in identity work, AI-enabled capabilities may support how access risks are handled. The sources establish these as spending areas, not that every vendor’s implementation performs equally well.

A second part of the decision is people. The survey coverage reports that 91% of CISOs expected AI to make their teams more productive over the next 12 months, and 81% expected it to create demand for new cybersecurity roles and skills. That sits awkwardly with the simplistic claim that AI just removes jobs. The reported expectation is transformation: repetitive tasks may shift, while teams need skills to assess outputs, investigate anomalies, and make decisions.

So ask whether a proposed tool changes a real workflow. Does it help an analyst get to a defensible investigation faster? Does it fit the identity process the team actually runs? Can staff recognize when the system is wrong? A good pilot should answer those questions in the environment where the work happens, not in a vendor demonstration.

The value problem is real

Security returns are unusually difficult to measure because their most important benefit is often an event that never happens: a breach, loss, or disruption. Several controls may contribute to stopping the same attack, making it difficult to assign credit to one product. AI doesn’t solve that accounting problem. It can make it worse if a team counts activity—alerts processed, summaries generated, tasks automated—as though activity were equivalent to risk reduced.

That’s why “wait for perfect ROI” is too rigid, but “buy because attackers use it” is too loose. If the exposure is urgent, a limited investment can be rational before a long-term business case is settled. The burden then shifts to the buyer: define what the pilot is meant to improve, compare it with a baseline, and decide in advance what result would justify expansion.

Use measures tied to the use case. For an operations workflow, examine whether the tool changes investigation time or analyst workload without degrading the quality of decisions. For identity work, choose a measure that reflects the actual access-risk process. These are evaluation recommendations, not outcomes guaranteed by the research. Keep a human review step where consequences matter, and record failures as carefully as wins.

Buy narrowly, learn quickly

A sensible approach is to separate the strategic question from the purchase question. It may be strategically prudent to understand AI’s role in both attack and defense. That does not mean every security team needs a broad deployment now. Start with a bounded problem, a defined owner, known data inputs, and a way to compare performance before and after. If the tool doesn’t improve the task—or creates risks the team can’t manage—don’t scale it merely because the category is fashionable.

Budget ownership deserves attention too. The survey figures show AI spending can sit in a dedicated security line, within general security spending, or in other organizational budgets. CISOs should make sure that split doesn’t obscure who evaluates the security impact, who oversees access and data, and who is accountable when the system produces a bad recommendation.

NIST’s AI Risk Management Framework offers a useful organizing reference. NIST describes it as voluntary guidance intended to help organizations incorporate trustworthiness considerations into the design, development, use, and evaluation of AI systems. It isn’t a purchase checklist or a promise of safety. But its lifecycle framing is a reminder to assess the system as it is used and evaluated, not just when it is bought.

Is investing before proof the right move?

Sometimes. A CISO facing a credible operational need may reasonably fund a small, controlled test before the evidence is mature. But urgency should buy learning, not a blank cheque. The case for broader investment should strengthen only when the organization can show that a tool addresses a defined need, works with its people and processes, and produces a meaningful improvement without creating unmanaged risk.

The market signal is clear: AI is competing for cybersecurity dollars even as overall budgets remain constrained. The value signal is less clear. Treat that gap honestly. Invest where the need is specific, measure what matters, and keep the option to stop. That is not hesitation; it’s how a security team makes a bet it can defend.

is ai in cybersecurityis ai in cybersecurity