Behind ADT-4: Architecture, Reasoning System, and Safety Layers

Introduction

ADT-4 was created for environments where accuracy, stability, and traceability matter. Unlike general models that try to maximize breadth, ADT focuses on depth: understanding security signals, reasoning about risk, and making decisions that can be audited and trusted.

This post explains the architecture that enables these capabilities.

The Core Intelligence Architecture

ADT’s architecture is structured around a few foundational ideas.

Intelligent Signal Understanding

Security events are interpreted as connected narratives. Instead of processing logs as isolated lines, ADT reconstructs context:
who acted, when it happened, what changed, and why it matters.

Multi Step Reasoning

ADT evaluates threats and compliance gaps through multiple steps, refining its interpretation as additional evidence appears. It does not collapse complex decisions into a single guess. The reasoning process maintains clarity even under pressure.

Long Horizon Context

Investigations often span hours or days. ADT uses retrieval and structured memory to preserve context across extended timelines, preventing drift and maintaining consistency.

Adaptive Alignment System

ADT is designed to behave predictably, especially in high stakes environments. Its alignment system ensures actions remain within boundaries that organizations trust.

Safety Gates

Before ADT carries out high impact defensive actions, the system runs checks on intent, source, and contextual risk. This prevents unintended escalation and guarantees that responses stay within approved policies.

Context Guardians

A set of internal monitors continuously check reasoning steps for stability and coherence. If a conclusion falls outside expected patterns, the system requests additional evidence or shifts to a conservative posture.

Evidence First Principle

ADT does not act without traceable justification. Every major decision is accompanied by a chain of reasoning, contextual summary, and verifiable evidence. This ensures transparency for auditors and executives.

How ADT Interprets Risk

ADT’s risk engine evaluates events using a combination of structured logic and model based reasoning.

Threat Likelihood

ADT scores the probability that an event or pattern is malicious using behavioural signals, historical baselines, and global intelligence.

Impact Evaluation

The system assesses what could happen if the signal represents an active threat. This includes data exposure, lateral movement potential, and operational disruption.

Policy and Compliance Weighting

ADT maps evidence against applicable frameworks. For regulated industries, this means decisions are informed by real obligations, not guesswork.

Safety Layers Built For Institutions

Institutions need a model that behaves the same way every time. ADT-4 includes several layers that protect consistency.

Boundary Enforcement

All defensive actions pass through controlled boundaries that can be audited, tuned, and versioned. No autonomous action bypasses organizational policy.

Stability Under Load

ADT maintains reasoning integrity even when receiving large volumes of events. The architecture is designed to degrade gracefully under overload, never unpredictably.

Transparent Failure Handling

When ADT cannot determine a safe conclusion, the system defaults to clarity: it explains the uncertainty, logs the requirement for human review, and provides relevant evidence.

Built For Real Infrastructure

ADT integrates into cloud, hybrid, and on premise environments using two paths:

Agentless Ingestion

Direct log and signal submission into ADT’s reasoning engine. Ideal for cloud and application workloads with minimal deployment effort.

Lightweight Agents

Optional agents for organizations that need deeper visibility into servers, containers, and legacy systems.

Both paths feed into the same intelligence core, ensuring consistent interpretation regardless of the integration method.

A message from the CEO

“ADT’s architecture is built around a simple idea. Systems that protect institutions must think clearly, act reliably, and explain themselves. ADT-4 reflects that belief at every level of its design.”
David Idris, Founder and CEO