Three numbers from the last twelve months should change how you think about your security operations centre. CrowdStrike’s 2026 Global Threat Report found that attacks by AI-enabled adversaries rose 89% year on year. They also reported that the average time an eCrime actor needs to move from their first compromised host to a second system fell to 29 minutes, with the fastest breakout they observed taking just 27 seconds.
Now set that against the defensive side. The 2026 survey run by Prophet Security of 300 CISOs, SOC leaders and practitioners, found that large enterprises generate an average of 3,181 alerts a day. Around 40% of those alerts are never investigated.
One side of the contest is accelerating with AI. The other is still queuing alerts for people to read. If you are accountable for detection and response in your organisation, this mismatch is the risk. It is already in your environment, whether or not it has reached your board pack.
In this blog article Cyberfort security experts explore why many Security Operation Centres are failing to keep up with AI threats and offer practical advice on actions to take to improve AI threat detection, response and triage.
The priority is speed and signal, not headcount
The usual reaction to a stretched SOC is to hire. It is understandable, but it no longer works. Alert volumes grow faster than any recruitment plan, and the analysts you do recruit are being asked to win a race against adversaries who now use AI to automate reconnaissance, craft convincing lures and move laterally in minutes.
So, the focus must change. The question is no longer “how many analysts do we have?” It is “how quickly does a genuine threat reach a person who can act on it, and how much of our expensive human attention is wasted on everything else?”
This gives Security teams three priorities which they should be focusing on.
1. Compress detection and response time.
If the average breakout is 29 minutes, a detection process that depends on an analyst picking up an alert, gathering context from four tools and escalating it is already behind. Your target is not “faster triage”. It is containment inside the attacker’s window.
2. Protect analyst judgement.
Your people are your scarcest asset. Every hour spent closing obvious false positives, copying indicators between consoles or writing the same case notes is an hour not spent on the intrusion that matters. The aim of AI in the SOC is not to replace that judgement. It is to make sure it is spent where it counts.
3. Close the visibility gap.
When roughly 40% of alerts go un-investigated, your business has a blind spot you chose by default. Some of those alerts will be noise. Some will not. You cannot tell which is which unless something is looking at all of them.
These three priorities decide whether your detection capability is keeping pace or quietly falling behind. Technology choices follow from them.
Why the gap has opened so fast
When looking at why AI powered attacks have opened up response gaps so fast in the past year, several forces have combined, and each makes the others worse.
Attackers are industrialising with AI.
CrowdStrike’s finding that AI-enabled adversary activity rose 89% is not a story about exotic nation-state tooling. It reflects AI becoming a standard accelerant. Faster phishing at scale, quicker vulnerability research, more convincing impersonation and more efficient post-compromise activity. Adversaries do not need to be more skilled. They need to be quicker and more numerous, and AI delivers both.
Exploitation windows are collapsing. Weekly or monthly patch and review cycles were designed for a slower adversary. Your detection layer now has to assume that exploitation can follow disclosure almost immediately and catch the behaviour rather than rely on a signature being published first.
Your telemetry has outgrown your tooling. Cloud workloads, identity platforms, SaaS applications, endpoints and now AI tools and agents all produce signals, often in separate consoles with no shared context. Adding a point product for each new risk has multiplied alerts without multiplying understanding. An analyst looking at three low-confidence alerts from three tools cannot easily see that together they describe one intrusion.
Alert fatigue is a security failure, not a morale issue. It is tempting to file this under staff wellbeing. It is more serious than that. When analysts face thousands of alerts a day, triage shortcuts become routine. Genuine signals get closed alongside the noise because the previous twenty were false positives. A control that is in place but not looked at does not protect you.
Confidence and capability have drifted apart. Many organisations report that their SOC is “covered” because it has the right platforms in place. Coverage is not the same as outcomes. Mean time to detect, mean time to respond and the share of true positives that reach a human are the measures that show whether the SOC works. Raw alert counts and tool inventories do not.
What links all of this is that the attacker’s cost of operating has fallen, and the defender’s cost of reviewing has not. Closing that gap needs a change in how the SOC processes signal, not just how many people sit behind it.

An AI enabled SOC that earns its keep
A credible response combines detection engineering, automation and AI-assisted analysis. It is built around your environment and delivered as an operating capability, not a one-off tool deployment. So, what should organisations be looking for from their SOC when trying to keep up with the influx of AI generated attacks? From our experience at Cyberfort there are 8 key areas which need to be looked at:
1. Start by measuring where time goes
Before you automate anything, establish a baseline. Which alert types consume the most analyst hours? Which detection rules produce the most false positives? How long does a typical investigation take from alert to decision? This gives you evidence to prioritise against and a way to prove improvement afterwards.
2. Improve detection quality, not just quantity
Many SOCs carry years of accumulated rules that nobody has reviewed. Detection improvement means auditing that content against the techniques attackers use now, retiring rules that add noise without value, and building behavioural detections for activity that signature-based approaches miss – unusual identity behaviour, abnormal data movement, suspicious use of legitimate tools. With AI-enabled adversaries varying their traps and tooling constantly, behaviour-based detection ages far better than indicators of compromise.
3. Use AI to triage and investigate at machine speed
This is where AI earns its place on the defensive side. AI-assisted triage can enrich every alert with threat intelligence, asset criticality and user context. It can correlate related signals across tools, suppress clear duplicates, summarise what happened in plain language and recommend a next step, all before a human opens the case. The aim is that analysts start each investigation with the picture already assembled rather than spending the first fifteen minutes building it. Be disciplined about this, though. AI in the SOC should be transparent and auditable. Your analysts need to see why a decision was suggested, and you need to be able to evidence it to an auditor or regulator. Treat AI-driven triage as a supervised capability with defined confidence thresholds, human approval for higher-impact actions and regular review of its decisions. It is not a black box you trust blindly.
4. Automate the response steps that are well understood
Isolating a compromised endpoint, disabling a suspicious account, revoking a session token, blocking a malicious domain and opening a ticket are repeatable actions. Where the trigger conditions are clear, automation can execute them in seconds, inside the 29-minute window that matters. Start with low-risk, high-frequency actions, prove they work, and extend automation as confidence grows. Keep human approval in place for actions that could disrupt the business.
5. Correlate across your whole estate
The attack chain rarely stays inside one tool. An effective design joins identity, endpoint, network, cloud and SaaS telemetry so that related events are presented as a single incident with a timeline. Correlation reduces alert counts and, more importantly, it exposes the intrusions that individual low-confidence alerts hide.
6. Secure the AI you are now defending with
AI tools and agents expand your attack surface as well as your defences. AI tools and development platforms are themselves being actively exploited. If your SOC uses AI, or your business is adopting it, those systems need the same threat modelling, access control and monitoring as any other critical asset.
7. Treat tuning as a continuous discipline
Threat behaviour changes monthly. Detection logic, automation playbooks and AI models configured for last year’s attacks degrade every week they go unreviewed. Build a regular cycle of review, testing and improvement, including exercises that check whether your detections would catch current attacker techniques.
8. Report on outcomes your board understands
Replace volume metrics with measures that show risk reduction: time to detect, time to contain, the proportion of alerts investigated, false positive rate and analyst hours returned to proactive work. These figures show whether the investment is working and give you a defensible story for your board.
For many organisations undertaking the 8 key actions above alone without specialist support is almost an impossible task, given that many IT and Security teams are already under resourced and do not have the in-house knowledge or expertise to implement successfully. In short – most organisations that try to use purely internal resources discover that the problem is not tooling but capacity and expertise.
Why you need a specialist MSSP partner, and what to do next
Detection engineering, SOC automation and AI-assisted analysis are distinct specialisms. They differ from running a SOC day to day, and they are being asked of a team that is already fully occupied. Your analysts cannot step away from the queue for six months to redesign it. Hiring detection engineers and automation specialists is slow and expensive, and competition for them is intense. Even when you succeed, a small internal team will see only a fraction of the attacker behaviour that a specialist provider sees across many environments every day.
A specialist managed security services partner closes those gaps in practical ways:
Cross-environment intelligence. A partner sees how attackers behave across many organisations, so detection content is tested against real tradecraft and refreshed continuously, not rebuilt from scratch in your environment alone.
Proven automation and AI practice. Playbooks, correlation logic and triage models that already work elsewhere can be adapted to your estate, with the governance and human oversight your risk team needs.
Capacity without compromise. You gain 24/7 coverage and engineering depth without diverting your own people from the work only they can do.
Integration, not replacement. The right partner builds on the platforms you already run. It improves what you have rather than forcing a costly rip-and-replace.
Accountability for outcomes. You can hold a partner to measurable service levels on detection and response time, not just on delivering a tool.
If your business is struggling to find the right solution to the AI attack volume problem the next steps should be:
- Review your own numbers. Find out how many alerts your SOC receives daily, how many are investigated, and how long it takes to go from alert to containment. If you cannot answer quickly, that is your first finding.
- Compare them to the attacker’s clock. Set your response time against a 29-minute average breakout. Be honest about the gap.
- Commission an independent assessment. Ask a specialist to review your detection coverage, automation maturity and AI readiness against current attacker behaviour, and to give you a prioritised roadmap rather than a product list.
- Start with the highest-value improvements. Typically, that means reducing noise, automating the most common response actions and adding behaviour-based detection for your most critical assets.
- Build for the long term. Agree a model where detection and automation are tuned continuously, measured against outcomes and reported to your leadership in terms of risk.
The adversary has already adopted AI, and it is saving them time. Your SOC can use the same technology to take that time back, provided you apply it with discipline, governance and expert support. Cyberfort’s MXDR service is built to do exactly that. Expert-led, security driven, and measured on how quickly real threats are stopped.
If you would like to know how your own detection and response compares with the attacker’s clock, the right place to start is a conversation with a security specialist at Cyberfort. To book your SOC review email us at [email protected] and one of our experts will be in touch.





















