Back to blog

Security is not just safe or dangerous

Cybersecurity's Biggest Lie Is That Everything Is Either Safe or Dangerous

NetSenX Team12 min read
ndrevidenceincident-response

Real risk is not a switch. It is a story that develops between signals

At 10.46 p.m., a company server began transferring a large amount of data to a cloud service. The connection was encrypted. The certificate was valid. The account had the correct permissions, and the service appeared on the organisation’s list of approved providers.

Was this a legitimate backup or the beginning of data theft?

A few minutes earlier, an administrator account had logged in from an address never previously seen in the organisation’s history. That still did not necessarily mean an attack. The administrator may have been working remotely, using a new network or carrying out scheduled maintenance.

At the same time, one workstation began scanning internal systems, while the data transfer continued for significantly longer than previous backups. The destination was legitimate, but the data centre was located in a different region than usual.

At what point does the situation become dangerous?

Or are we still looking at several legitimate events that happened to occur at roughly the same time?

Moments like this expose one of the most damaging simplifications in modern cybersecurity. The belief that every activity can be quickly and honestly assigned to one of two categories. Safe or dangerous. Allow or block. Attack or no attack.

Reality is rarely that convenient.

We want simple answers, but the real world does not provide them

People naturally seek certainty. Green creates reassurance. Red demands action. The board wants to know whether the company is safe, while an analyst wants to decide whether to close an alert or begin an investigation.

The problem is that an answer that looks precise is not always an answer that is true.

A new connection may be a legitimate integration or an attacker’s control channel. An unusual transfer may be a backup or exfiltration, meaning the unauthorised removal of information from an organisation. A regular packet may be a harmless device status signal or beaconing, meaning rhythmic communication between a compromised system and an attacker’s infrastructure.

A supplier connecting at night may be carrying out scheduled maintenance. The same activity may also involve a compromised account. A certificate change may be a routine renewal or evidence that the cryptographic protection of a connection has been altered without authorisation.

At the level of a single event, each scenario may look almost identical.

Meaning appears only when we understand the history, behaviour, relationships and context.

How often do the systems inside your organisation answer a question with complete certainty when the honest answer should be that we do not yet know?

A single signal rarely tells the whole story

Imagine an employee’s computer connecting to a new domain.

A system based on a simple rule may check whether the domain appears on a threat list. If it does not, the activity may be treated as safe. If it does, an alert is generated.

But what happens when the attacker uses newly created infrastructure, a legitimate cloud provider or a domain that has not yet developed a malicious reputation?

In that situation, the device’s behaviour may matter more than the name of the destination. Has the computer communicated with it before? Are connections appearing at perfectly regular intervals? Does the amount of transferred data match the device’s role? Are similar computers behaving in the same way? Was there earlier network scanning, an unusual login or an attempt to access another system?

None of these elements alone may confirm an attack. Together, however, they can create a picture that no single dashboard would reveal.

Effective threat detection should therefore combine timing analysis, comparison with peer devices, network relationships, historical activity and changes in data volume. It is not enough to ask whether an event matches a known signature. We must ask whether the current behaviour still makes sense for this particular device, account, process and business situation.

Do your security tools see individual alerts, or do they understand the story those alerts tell together?

Behaviour is often more important than the appearance of an attack

A signature describes something that has already been identified. It may be a known file, communication pattern, address or fragment of malicious code.

These mechanisms remain necessary and continue to provide an important layer of protection. Their limitation becomes clear when a threat changes its appearance.

A new malware variant may use different code, a different name and a new communication address. It still needs to perform certain actions. It must establish contact, explore the environment, reach additional resources, maintain access or transfer information.

Appearance can be changed relatively easily. Behaviour is much harder to hide.

This is why behavioural network analysis does not focus only on what a packet contains. It observes who communicates with whom, when communication occurs, how often it happens, how regular it is, how much data moves and how relationships between devices change.

NDR means Network Detection and Response. It is the detection and analysis of threats visible in network traffic. This layer observes real communication between computers, servers, devices, cloud services and suppliers.

It does not replace firewalls or endpoint protection. It adds the perspective that is often missing when every tool sees only one fragment of the environment.

How many threats inside your organisation could look normal to one system but completely different when combined with network traffic and behavioural history?

Security is not binary because context is not binary

Imagine two employees downloading the same file.

The first uses a company laptop during normal working hours from a location used every day. The document relates directly to the employee’s project, and similar activity has happened many times before.

The second uses an account that has been inactive for several months. The login occurs after midnight from a new location. The account then downloads files from several departments with which it has never previously interacted.

Technically, both employees performed the same action.

Is the level of risk the same?

Clearly not.

This is why a security system should not evaluate activity only against a single rule. It should assess the degree of deviation, the strength of the evidence, the relationships between signals and the remaining uncertainty.

Multi-valued logic moves beyond the simple division between true and false. In practice, it allows a system to recognise intermediate states. Activity may be consistent with normal behaviour, new but justified, worthy of observation, highly suspicious or supported by strong evidence of a real incident.

That is how an experienced analyst reasons. A decision is not based on one colour. Information is combined, and the assessment changes as new evidence appears.

Technology should be able to do the same.

Can a system that must immediately answer safe or attack honestly describe a situation in which the evidence is still developing?

A risk score of 92 percent may mean almost nothing

Many security dashboards present impressively precise results. Risk at 87 percent. Confidence at 94 percent. Severity at 9.2 out of 10.

These figures look scientific. They create a feeling of control and make it easier to sort an alert queue.

But what do they actually mean?

How much data was used to produce the score? How does the model behave in a new environment? How often did a similar result lead to a false alarm? Was uncertainty measured, or simply hidden behind a number with two decimal places?

A risk score without information about the quality of the evidence may be little more than a guess presented with mathematical precision.

A mature system should show not only the result, but also the limits of its own confidence. It should explain which factors increased the assessment and how strongly the behaviour differs from its baseline. It should also be capable of saying that there is not enough data to support a reliable conclusion.

That is not weakness.

It is a sign of maturity.

Would you trust a system that always claims to know the answer, or one that can clearly state when the available evidence is still insufficient?

A false alarm is not an innocent mistake

Every security system can generate false alarms. The real problem begins when their number is no longer under control.

SOC means Security Operations Center, the team responsible for monitoring and investigating security events. Such a team may receive hundreds or thousands of warnings every day. When most prove irrelevant, analysts lose time, concentration and trust in their own tools.

A real attack does not always need to remain undetected.

It may be detected and still disappear inside the queue.

A good system should therefore not be evaluated only by the number of anomalies it identifies. It also matters whether it controls alert quality and limits false positives.

Conformal prediction is a statistical method that can create verifiable uncertainty boundaries under clearly defined assumptions. It does not promise zero mistakes and does not guarantee detection of every possible threat. It can, however, help control an upper boundary for the proportion of false alarms and allow independent verification of whether that level is genuinely being maintained.

The most important principle is simple. When there is not enough data to confirm a result honestly, the system should not manufacture a number simply because the dashboard expects one.

Insufficient data is also information.

Why do we so readily accept marketing claims about high accuracy, yet so rarely ask whether the customer can independently verify the quality of the alerts being generated?

One event may be harmless while the sequence is not

Let us return to the server transferring data to the cloud.

The transfer itself may be legitimate. The new login may also be legitimate. Even network scanning may result from normal administrative work.

But what if the sequence began with a login from a new location, followed by scanning, access to several internal systems and finally a large transfer to a destination the organisation had never previously used?

Each event considered separately may receive a low or medium risk assessment.

The complete sequence may reveal a developing attack chain.

Reconnaissance means exploring the network and identifying available systems. Lateral movement means an attacker moving between devices after gaining the first foothold. Exfiltration means removing information from the organisation.

The company does not need three unrelated alerts on three different screens. It needs one coherent story showing where the activity began, how it developed, which devices and accounts became involved and where the path was leading.

This is why graph analysis, network maps, incident grouping and attack-chain reconstruction matter. Risk often does not exist inside one element. It exists in the relationship between several elements.

After an incident, will your team see one chronological story, or spend several days manually comparing exports from different tools?

The explanation matters more than the colour

A high-risk alert should answer the question of why it exists.

It may explain that a device contacted a new external destination, the volume of data was several times higher than usual, the activity began outside normal operating hours and peer devices showed no comparable behaviour.

It may also show that network scanning, a change of technical account or an attempt to reach a system containing sensitive information occurred earlier.

This attribution layer shows which factors influenced the risk assessment. A natural-language explanation helps people who do not analyse network packets every day understand the event. A decision trail makes it possible to reconstruct later why the system produced a particular recommendation.

This matters to the analyst who needs to assess the event quickly. It also matters to the board, auditor, insurer and regulator.

A black box can produce a score.

It cannot create trust.

Should the person responsible for the company make a decision based on a red colour, or on understandable evidence showing what actually changed?

Not every uncertainty should end in an automatic block

The simplest response to uncertainty may appear to be blocking everything that deviates from the norm.

In theory, this sounds safe. In practice, it can interrupt production, sales, logistics, customer service or a financial process.

New behaviour does not always mean a threat. Companies introduce applications, update systems, change suppliers and redesign processes. If every deviation triggers an automatic block, the security platform begins fighting the business it was meant to protect.

A responsible approach should therefore begin with continuous monitoring.

The system observes, classifies, connects signals, explains the risk and presents a recommendation. Enforcement can remain with firewalls, endpoint protection, event-management platforms and responsible human decision-makers.

EDR means Endpoint Detection and Response. It monitors activity on devices such as computers and servers. SIEM means Security Information and Event Management. It collects and correlates security logs. SOAR means Security Orchestration, Automation and Response. It coordinates and automates agreed response procedures.

The analytical layer does not need to replace those systems. It should provide the context, evidence and explanation required for a responsible response.

A lack of evidence should not automatically lead to a high-impact action. At the same time, a lack of absolute certainty should not mean that the organisation ignores a developing risk.

Intelligent control must operate between these two extremes.

How should organisations balance the risk of missing a threat with the risk of interrupting the business through an incorrect response?

The board does not need certainty. It needs an honest picture of risk

A chief executive does not need to know that the security platform generated 638 alerts last month.

Leadership needs to know whether the business was genuinely exposed, which processes may have been affected, whether data left the organisation, how reliable the assessment is and what action is recommended.

A good report should distinguish a confirmed incident from activity that requires further observation. It should show which signals are strong, which are weak and what remains unknown.

This is particularly important for cybersecurity in Poland and across Europe, where organisations increasingly need not only to detect incidents, but also to document how those incidents were assessed and handled.

NIS2 is the European Union directive on the security of network and information systems. DORA means Digital Operational Resilience Act and applies to the financial sector. GDPR means General Data Protection Regulation and governs personal data protection as well as obligations connected with breaches.

No tool automatically makes an organisation compliant with these regulations. It can, however, provide supporting evidence, decision trails and workflow assistance that help demonstrate what the organisation observed, how the situation was assessed and why specific actions were taken.

Does your board receive an honest view of risk, or only a colourful representation of technical certainty that nobody can defend?

Uncertainty is not the greatest threat

Uncertainty is a natural part of cybersecurity.

We do not immediately know whether a new connection is an update, a configuration error, supplier activity or the first stage of an attack. We do not know whether an increase in data transfer means a backup, a process change or a leak.

The greatest threat is not that the system does not know the answer from the first second.

The greatest threat is a system that pretends it does.

Mature cybersecurity should not promise a world without uncertainty. It should measure it, explain it and update the assessment as new information appears.

It should observe behaviour, connect relationships, compare devices, detect anomalies and preserve evidence. It should reduce false alarms without pretending that errors can be eliminated entirely.

Cybersecurity’s biggest lie is therefore not necessarily that someone is deliberately misleading customers.

It is the comfortable illusion that a complex digital environment can always be described with one colour.

It cannot.

What can be built instead is a system that shows why risk is increasing, how strong the evidence is and what the organisation should do next.

Does cybersecurity in your company measure uncertainty or hide it?

Would you rather use a system that gives an immediate and simple answer, or one that presents the real picture even when the answer is not yet convenient?

Perhaps the most important change in cybersecurity will not be faster detection. It may be the courage to stop pretending that every risk is black or white from the beginning.

Further reading: what NetSenX detects and how each alert explains itself, NDR vs EDR vs SIEM for teams without a SOC and how NetSenX compares with other NDR vendors.


Written by the NetSenX Team at TriStiX S.L. Questions: info@netsenx.com.

NetSenX Team

TriStiX S.L.