DEV Community

Saurabh Kumar
Saurabh Kumar

Posted on

Beyond Block or Allow: Designing Adaptive Defense in CIDS.

Most security systems eventually have to answer a simple question:

Allow or block?

That model is useful.

But I think there is an interesting space between those two decisions.

What if the system isn't completely certain that an activity is malicious?

What if the behavior is suspicious, but blocking immediately would create unnecessary friction?

And what if the system has enough evidence to know that something is wrong, but instead of simply terminating the interaction, it can safely contain it and gather more defensive intelligence?

These questions led me to one of the core ideas behind CyberMoranda CIDS:

Adaptive Defense.

Detection Is Not the Same as Response

CIDS separates these two decisions.

Behavior

Detection

Risk + Confidence

Policy

Response

This distinction is important.

The detection engine should not decide:

"I found something suspicious, therefore BLOCK."

Instead, it should provide evidence.

The policy engine decides what that evidence should mean for the environment.

A Response Spectrum

The response architecture I'm working toward looks like:

┌── Allow

├── Monitor

Risk + Confidence ──┼── Slow / Challenge

├── Contain

└── Deception

These aren't simply five fixed risk levels.

The actual response should depend on context, confidence, asset sensitivity, policy, and observed behavior.

Why Not Block Immediately?

Imagine a request pattern that is unusual but not conclusive.

Risk: HIGH
Confidence: LOW

Immediately blocking could be a bad decision.

Instead, the system might increase observation:

Suspicious activity

Increase monitoring

Observe next actions

Update context

Re-evaluate risk

This gives the system more evidence before taking a stronger action.

That is the basic idea behind adaptive response.

Dynamic Friction

One concept I am particularly interested in is dynamic friction.

Instead of applying the same delay to every suspicious request, the response can be influenced by the current context.

For example:

Low concern

Normal response

Moderate concern

Additional observation

High concern

Controlled friction / challenge

Very high confidence

Containment

The exact implementation and thresholds need to be validated through testing.

I don't want arbitrary numbers to become part of the architecture simply because they looked good in an MVP.

Deception as a Defensive Layer

Deception is another part of the CIDS design.

The objective isn't to attack the attacker.

It is to create a controlled environment where suspicious activity can be observed without exposing the real application or sensitive resources.

Conceptually:

Suspicious Session

Policy Evaluation

Is Deception Appropriate?

Controlled
Deception

Observe + Record

The real application remains separated from the deceptive environment.

What Could Deception Provide?

A controlled deception environment could help collect information such as:

  • Requested paths
  • Request sequences
  • Timing
  • Interaction patterns
  • Authentication attempts
  • Tool/fingerprint signals
  • Session behavior
  • Commands or inputs attempted within the decoy

The goal is defensive visibility, not retaliation.

Honeytokens

Another concept that fits naturally into this architecture is the use of honeytokens.

A honeytoken is an intentionally placed piece of data designed to generate a signal when accessed or used unexpectedly.

For example:

Normal application

Hidden defensive marker

Unexpected access

Security Signal

Risk Evaluation

A honeytoken event can become one piece of evidence within the broader behavioral model.

It should not automatically mean:

Honeytoken → BLOCK

Instead:

Honeytoken
+
Session Context
+
Previous Behavior

Risk Evaluation

Again, context matters.

Deception Should Not Become a Trap for Everyone

This is an important engineering constraint.

A defensive deception mechanism must be isolated.

It should not:

  • expose real credentials
  • provide access to production systems
  • become an uncontrolled pivot point
  • interact offensively with external systems
  • create unnecessary risk for legitimate users

The deception environment should be treated as an isolated security boundary.

Feedback Changes the System

Another important part of the architecture is feedback.

Suppose CIDS makes a decision:

Behavior

Risk

Contain

Later, an analyst determines that the behavior was legitimate.

That information shouldn't disappear.

Conceptually:

Detection

Decision

Outcome

Analyst Feedback

Policy / Detection Tuning

This doesn't necessarily mean blindly training an AI model from every analyst action.

It means the system should have a structured way to learn from operational outcomes.

The Bigger Architecture

This brings the different CIDS components together:

Network / Host Telemetry

Event Normalization

Session Context

Behavior Analysis

Signal Engine

Risk + Confidence

Policy Engine

Adaptive Response
/ | \
Allow Monitor Contain

Deception

Evidence / Audit

Feedback

The dashboard is simply a window into this system.

The real product is the decision-making backend underneath it.

What I Am Trying to Build

I don't want CIDS to become another product that simply shows:

ATTACK DETECTED
IP BLOCKED

I want the system to answer a more useful set of questions:

What happened?

Why does it look suspicious?

How confident are we?

What has this session been doing?

What response is appropriate?

What did we learn from the response?

That is the direction behind the adaptive defense architecture of CIDS.

There is still a lot of engineering and validation ahead.

The architecture is evolving.

The prototype is only the beginning.

Observe. Understand. Adapt. Defend.

CyberMoranda CIDS — Think Before You Act.

Top comments (0)