Summary
- Securonix research involving 1,000 cybersecurity professionals found that 99.6% reported increased budgets for AI security automation.
- Although 87% reported some AI adoption, 53% favoured human approval within their overall cybersecurity automation strategy.
- The study identifies changing security operations roles, growing governance requirements and concerns about third-party AI visibility.
Large organisations are increasing investment in artificial intelligence for cybersecurity operations while retaining human oversight of consequential decisions, according to research examining how security teams are balancing automation with accountability.
A study commissioned by Securonix found that 99.6% of respondents had increased their budgets for AI adoption in cybersecurity automation during the previous year, while 53% favoured an overall approach in which people remain directly involved in approving automated decisions.
The research surveyed 1,000 cybersecurity professionals at organisations employing between 2,000 and more than 50,000 people across the UK, France, United States and Australia.
Although AI is becoming established within security operations, the findings reveal differences between the tasks organisations have automated and the level of autonomy they believe should be permitted.
Automation expands across security operations
Almost nine in ten respondents, 87%, reported using AI to some extent within cybersecurity automation. Some 96% considered cybersecurity automation important, while 63% described it as very important.
Adoption levels varied considerably. Approximately 16.4% reported that AI systems could dynamically coordinate and adapt security workflows with minimal human intervention, while 17.2% described their operations as fully autonomous and AI-driven.
Another 13% said their organisations did not use AI in cybersecurity automation.
These categories reflect respondents’ descriptions of their own operating environments. They do not establish how reliably the systems perform or whether their claimed level of autonomy has been independently assessed.
The distinction becomes important as security platforms move beyond analysing alerts and recommending actions towards carrying out changes within enterprise infrastructure.
An AI system that groups related detections or summarises evidence operates under different conditions from one authorised to disable an account, isolate a production server or alter network access controls.
The latter activities can reduce the time available to an attacker, but they can also interrupt legitimate business operations if the detection or automated decision is incorrect.
Organisations must therefore determine which actions can proceed automatically, which require approval and how an intervention can be reviewed or reversed.
Human oversight remains the preferred approach
Across their overall cybersecurity strategies, 53% of respondents favoured a human-in-the-loop approach, in which a person remains involved in approving decisions before execution.
For individual security operations centre functions, 48% reported using that model. A further proportion operated systems under human supervision rather than requiring approval for every action.
Both approaches retain human responsibility, although they differ in how decisions are authorised and when personnel can intervene.
The survey also found that 35% believed alert triage would always require human oversight, while 19% reported that the function was already fully autonomous within their organisations.
That difference illustrates the absence of a settled approach to automation even for relatively established security operations activities.
On average, 32% believed that functions such as alert triage and incident response would continue requiring human involvement indefinitely.
Approval arrangements need to reflect the consequences of the action involved. Automatically enriching a detection with additional information presents a different operational risk from revoking privileged access or terminating a process supporting a business application.
Where automated tools possess authority to act directly, audit records and clearly defined permissions become necessary for investigating decisions that produce unexpected results.
Security roles adapt to AI deployment
The research also identifies changes in the skills required to operate and govern security automation.
Thirty-five per cent of respondents reported retraining analysts for higher-value activities, including investigation quality, decision-making and threat interpretation. Another 33% identified emerging roles associated with AI governance, validation and oversight.
Meanwhile, 31% reported hiring additional specialists as existing cybersecurity responsibilities changed.
These figures do not establish a single workforce outcome for AI adoption. Automation may reduce repetitive work while increasing demand for personnel capable of reviewing system behaviour, validating findings and managing operational controls.
The emphasis also differs between security leadership functions. Chief information security officers placed greater importance on governance and accountability, while security operations leaders were more concerned with skills and workforce development.
Incident response leaders tended to prioritise speed and faster resolution. Cyber threat intelligence specialists placed greater emphasis on trust, validation and transparency.
Those differences can complicate investment decisions where departments expect different outcomes from the same technology.
Governance maturity remains difficult to measure
Ethical concerns were the most frequently identified source of unease about AI in cybersecurity, cited by 26% of respondents. Limited visibility into third-party tools and dependence on a single AI provider were each cited by 24%.
The concerns extend beyond the accuracy of automated detections. Organisations also need to understand how AI systems handle sensitive information, which external services process that data and how access permissions are enforced.
Some 42% of respondents described their AI governance as defined or managed, while 27% considered it optimised. Those assessments were self-reported and do not provide independent evidence that the associated controls are effective.
Respondents placed greater weight on operational outcomes than direct cost reduction when evaluating AI investment. Faster detection and response, reduced investigative workloads, improved analyst productivity and better detection quality were prominent measures of value.
Tool consolidation and cost savings ranked lower, indicating that the surveyed security professionals generally viewed AI as a means of improving operational performance rather than simply reducing expenditure.
The study was conducted during summer 2026 and covers large organisations in four countries, including selected regulated and infrastructure-intensive sectors. Its findings should not be extrapolated to smaller businesses or treated as representative of every national cybersecurity workforce.
Within that population, the combination of widespread adoption and continued demand for oversight indicates that organisations are expanding AI’s operational responsibilities while developing the controls needed to govern its decisions.





