New

AI systems are advancing faster than human judgement can keep up

Safety of AI identifies where human judgement is replaced, where reliance forms, and where risk becomes invisible

Framework Overview

Where AI Systems Break Down

AI risk begins when adoption outpaces human judgement.

These patterns define how that risk emerges in practice.

  • System Decisions

    • Human Review Required

    • AI Dependency Risk

      Confidence: 72%

    • Human-in-the-Loop Failure

      Confidence: 85%

    • Judgement Displacement

      Status: Escalation required

    • Second-Order Effects

      Status: Human override missing

    • Automation Drift

      Status: Human override missing

  • System Decisions

    • Human Review Required

    • AI Dependency Risk

      Confidence: 72%

    • Human-in-the-Loop Failure

      Confidence: 85%

    • Judgement Displacement

      Status: Escalation required

    • Second-Order Effects

      Status: Human override missing

    • Automation Drift

      Status: Human override missing

AI Dependency Risk

When judgement shifts to systems

Judgement Displacement

Reliance Patterns

Human-in-the-Loop Failures

Over-trust in AI systems gradually degrades human decision-making — long before obvious failure occurs.

Human oversight often fades gradually — not suddenly.

Automation Bias

Trust Decay

System Review Summary

Human oversight reduced across workflow actions

|

Add document

Auto-Approved

Review Skipped

94% Verified

System Review Summary

Human oversight reduced across workflow actions

|

Add document

Auto-Approved

Review Skipped

Automated Workflow Pipeline

Auto-Routed

Low Oversight

Normalised

Review

Process

Approved

Automated Workflow Pipeline

Auto-Routed

Low Oversight

Normalised

Review

Process

Approved

Second-Order Consequences

Exploring the behavioural, cultural, and organisational effects that emerge once AI systems become normalised.

Behavioural Drift

Responsibility Gaps

Explore The Framework