The Most Dangerous Space in Cybersecurity Is the Space Between Known Threats
Hatched by Honyee Chua
Aug 19, 2026
10 min read
1 views
58%
What if the next serious cyberattack does not resemble anything your defenses were trained to recognize?
Most security programs are built around categories. Malware, phishing, credential theft, ransomware, insider risk. Each category is useful because it gives defenders a name, a playbook, and a list of indicators to search for. But categories also create a hidden weakness: they encourage us to defend against what has already become legible.
The hardest threats often appear in the gaps between categories. They borrow the appearance of legitimate software, combine familiar techniques in unfamiliar sequences, or exploit a business process that no existing detection rule describes. To understand this problem, it helps to look at an unexpected analogy: the generation of moving images from still ones.
A powerful image model can begin with one visual concept and end with another, producing a sequence that gradually transforms one into the next. The important capability is not merely recognizing known objects. It is interpolating through the space between them. That same idea offers a new way to think about cybersecurity. Defenders should not only ask, “Can we detect this threat?” They should also ask, “What dangerous behaviors can emerge as an attacker moves between familiar states?”
That shift changes how we model risk, test systems, and organize collective expertise.
The blind spot created by categories
A classification system is a compression device. It takes a complicated reality and assigns it to a manageable label. This is indispensable in security. An analyst cannot investigate every packet, login, file, and process as if it were a unique philosophical event. Labels turn a flood of observations into patterns that humans and machines can process.
The problem begins when the label becomes more real than the behavior it represents.
Imagine a company that has excellent defenses against phishing. It filters suspicious links, trains employees to inspect domains, and requires multifactor authentication. An attacker may respond by abandoning the classic phishing email. Instead, the attacker compromises a trusted vendor, persuades an employee to approve an ordinary support request, and uses a legitimate remote administration tool to access a privileged environment. No single step looks like the organization’s mental image of phishing. Yet the sequence produces a similar outcome: unauthorized access through the manipulation of trust.
The threat has not disappeared. It has moved through the space between established categories.
This pattern appears everywhere. A stolen password is an identity problem. An unusual login location is an anomaly. A newly installed administration tool is a software event. A request for sensitive data is a business process issue. When these signals are examined separately, each may seem tolerable. When connected, they reveal an intrusion.
Security failures often occur not because a system cannot recognize known threats, but because it cannot recognize transitions between normal states.
That distinction matters. A static detector looks for an object: a malicious file, a suspicious domain, a known exploit. A transition oriented defender studies movement: a user who shifts from reading documents to creating accounts, a service account that moves from one application to an unrelated system, or a finance workflow that suddenly changes its approval rhythm.
The object is familiar. The path is not.
Interpolation is where novelty appears
In visual generation, the interesting result is often not the starting image or the ending image. It is the sequence between them. A human may understand a dog and a bicycle as separate concepts, but a generated transition can create a continuous visual world in which one gradually becomes the other. The model must represent relationships, not merely inventories.
Cybersecurity needs a similar representation of relationships among behaviors.
Consider a simplified attack progression:
- An employee authenticates from a normal location.
- The employee accesses a familiar cloud application.
- A new browser session appears.
- A permission is granted to an unfamiliar integration.
- Data begins moving to an external service.
- The employee’s account initiates actions outside their normal role.
At the beginning, the activity is ordinary. At the end, it is clearly dangerous. The central security question is not simply which event is malicious. It is when the trajectory became unsafe.
This suggests a useful model: represent an organization as a behavioral space rather than a list of assets. Each point in that space describes a state of the environment: who is acting, what they can access, which tools they use, what data is moving, and which business process is underway. Normal operations occupy regions of this space. Attacks often begin in those regions, then travel along unusual paths.
A secure system therefore needs to monitor both position and velocity.
Position asks: Where is the user, device, application, or process right now?
Velocity asks: How quickly and in what direction is its behavior changing?
A privileged administrator may legitimately access many systems. Their position is therefore not automatically suspicious. But if that administrator suddenly creates a new account, changes authentication settings, accesses payroll data, and exports archives within fifteen minutes, the velocity and direction of change deserve attention.
This is analogous to noticing not just that an object exists in an image, but that it is transforming rapidly into something else.
Why security communities matter more than perfect detection
No organization can precompute every possible transition an attacker might make. There are too many technologies, workflows, configurations, and human habits. A model trained only on yesterday’s examples will always struggle with tomorrow’s combinations.
This is why security communities are not merely places to exchange alerts or troubleshoot products. At their best, they function as collective pattern recognition systems.
One practitioner sees an odd authentication sequence. Another notices a similar event in a cloud environment. A third recognizes that the behavior resembles an older intrusion technique, but with a new tool. Individually, each observation is incomplete. Together, they outline a previously unnamed region of the threat landscape.
The value of a community is not that it eliminates uncertainty. It helps people compare partial maps.
This has an important implication for how organizations should share information. A bare indicator, such as an IP address or file hash, is useful but fragile. Attackers can replace it quickly. A behavioral description is more durable. “An external integration was granted access, followed by an unusual data export from a service account” teaches defenders how to recognize a family of transitions, including variants that have never been observed before.
The most valuable security knowledge therefore has at least three layers:
- The artifact: What domain, file, account, process, or tool appeared?
- The behavior: What did it do, and what changed before and after it appeared?
- The transition: How did the system move from an ordinary state toward a dangerous one?
Communities become much more powerful when discussion moves from the first layer to the third. The goal is not simply to circulate objects of suspicion. It is to improve everyone’s ability to see motion.
A threat intelligence system that records only what an attack looks like is building a museum. A system that records how normal activity became dangerous is building a navigation chart.
The practical challenge: detecting unsafe movement without paralyzing work
If every change is treated as suspicious, security becomes unusable. Businesses are dynamic. Employees change roles, applications are deployed, vendors are added, and legitimate emergencies produce abnormal behavior. A system that responds to every transition with a hard block will generate so many interruptions that people will work around it.
The answer is not to demand perfect certainty. It is to design graduated friction.
A low risk transition may receive no response. A moderate transition may trigger additional verification, a notification to the user, or temporary limits on data access. A high risk transition may require human approval or isolate the relevant account. The response should reflect not only the current event, but also the accumulated direction of travel.
For example, a new device login may be harmless. A new device login followed by a password reset, a change to multifactor settings, and an export of customer records is a different problem. Each event increases the probability that the account is being controlled by someone else. A sensible system raises friction progressively rather than waiting for a single unmistakable indicator.
This approach can be implemented with a simple concept: the transition budget.
Every identity, device, and service has a normal range of changes it can make within a period. The budget is not a fixed number of actions. It is a measure of how much behavioral distance the entity can travel before requiring explanation. A developer may have a large budget for code repositories but a small budget for financial records. A payroll service may have a narrow set of expected actions but a high sensitivity level.
When activity consumes the budget quickly, the system asks better questions:
- Is this change consistent with the person’s role?
- Does it follow an approved business process?
- Did several independent controls change at once?
- Is the behavior reversible if it turns out to be legitimate?
- What new capability did the transition create?
The final question is especially important. Attackers often seek not just access, but new options. A permission grant, a token creation, or a change in recovery settings can expand what becomes possible next. Security teams should prioritize transitions that increase an actor’s future power.
A new operating model for defenders
Thinking in transitions leads to several changes in daily security practice.
First, detection engineering should include path based scenarios. Instead of writing a rule only for “suspicious PowerShell” or “impossible travel,” teams should model sequences such as: unusual login, privilege change, access to a sensitive system, and outbound transfer. The sequence may be expressed through a rule, a graph query, or a human review workflow. The technical form matters less than the underlying question: what combination of ordinary actions becomes extraordinary?
Second, testing should include interpolation exercises. Security teams often run simulations that reproduce a known attack. Those are useful, but incomplete. A stronger exercise starts with a legitimate workflow and gradually alters one variable at a time: a different device, an unusual hour, a new integration, a change in data volume, or a new approval path. The team observes whether controls notice the drift before the scenario becomes clearly malicious.
Third, organizations should preserve context, not just alerts. An alert without history forces an analyst to reconstruct the path manually. Systems should make it easy to see what was normal for this identity, what changed recently, which capabilities were added, and what consequences followed. Context turns an isolated alarm into a comprehensible story.
Fourth, practitioners should share generalizable transitions. “This hash was malicious” has a short shelf life. “This attack began with a legitimate vendor invitation, then used delegated permissions to bypass the organization’s usual login pattern” remains useful even after the specific infrastructure disappears.
Finally, security leadership should measure the time between meaningful transitions, not only the time between compromise and containment. The earlier a team sees the trajectory changing, the more options it has. Prevention is often easiest before the attacker reaches the obvious stage.
Key Takeaways
-
Monitor behavior as a trajectory, not a collection of isolated events. Look for changes in direction, speed, privilege, data access, and tool usage.
-
Model the space between threat categories. Ask how phishing becomes identity abuse, how identity abuse becomes privilege escalation, and how privilege becomes data loss.
-
Share transitions, not merely indicators. Behavioral sequences remain valuable after domains, hashes, and tools have changed.
-
Use graduated friction. Increase verification and restriction as behavioral distance from the normal state grows, rather than blocking every anomaly.
-
Prioritize changes that expand future capability. New permissions, recovery methods, tokens, and administrative paths can be more important than the event that created them.
The real object of defense is change
Security has traditionally treated the organization as a collection of things: endpoints, accounts, applications, networks, repositories, and records. That inventory is necessary, but it is static. Attackers exploit the fact that real organizations are not static. They are processes, and processes are defined by change.
The most dangerous moment may not be when a malicious object enters the system. It may be when a familiar object acquires a new role, a normal account gains an unusual capability, or a routine action becomes the first step in a different sequence.
This reframes the defender’s task. The goal is not to recognize every possible threat in advance. That is impossible. The goal is to understand the organization’s normal behavioral landscape well enough to notice when movement through it becomes implausible, accelerating, or empowering.
The future of security will belong less to systems that know the most bad things and more to systems that understand how ordinary things become dangerous.
An attacker’s advantage often lies in the transition, because transitions are where labels fail. Defenders can reclaim that advantage by studying the paths between known states, learning collectively from partial observations, and treating change itself as a primary security signal.
The question is no longer only, “What is this?” It is also, “What is this becoming?”
Sources
Hatch New Ideas with Glasp AI 🐣
Glasp AI allows you to hatch new ideas based on your curated content. Let's curate and create with Glasp AI :)
Start Hatching 🐣