AI Existential Risk Lab

3D hypothetical escalation scenario visualizer

HYPOTHETICAL · NOT A PREDICTION
Scenario stage
1 · Development
Risk pressure
28%
Human control
72%
AI capability Critical infrastructure Escalating risk Safety barrier
DRAG TO ROTATE · WHEEL TO ZOOM · CONCEPTUAL 3D SCENARIO

A Hypothetical Escalation Path

This is not a forecast. It is a simplified scenario showing how several failures could compound if increasingly capable AI systems were given broad autonomy and access while oversight and resilience failed.
01

Rapid Capability Growth

AI becomes useful across research, software, industry, finance, information, and infrastructure.

02

Deep Dependency

Organizations delegate more decisions and operations to automated systems.

03

Control Failure

Objectives, incentives, or behavior diverge from what people intended.

04

Cascading Disruption

Interconnected systems amplify failures faster than institutions can respond.

05

Loss of Recovery Capacity

Human institutions lose enough coordination, infrastructure, or decision capacity to restore control.

06

Existential Outcome

In the extreme hypothetical case, irreversible global failure could threaten humanity's long-term survival.

What Could Increase the Risk?

Existential-risk arguments generally depend on multiple conditions occurring together. Capability alone does not imply catastrophe.

Misaligned Objectives

Goal Gap

The system pursues an objective that does not adequately represent human intent or constraints.

High Autonomy

Agency

The system can plan and act for long periods with limited human approval.

Broad Access

Reach

AI becomes connected to many consequential digital and physical systems.

Competitive Pressure

Race

Organizations or states may accept more risk when they believe competitors are moving faster.

Weak Evaluation

Blind Spot

Tests fail to reveal dangerous behavior before deployment or fail under new conditions.

Low Resilience

Cascade

Highly connected systems lack independent backups, isolation, or human recovery paths.

Where the Escalation Can Be Interrupted

The same scenario can be viewed as a map of intervention points. Independent safeguards can reduce the chance that one failure becomes a global cascade.

Capability Evaluation

Test new systems before granting more autonomy or access.

Human Authorization

Keep consequential actions behind meaningful human approval and accountability.

Access Boundaries

Limit privileges, isolate critical systems, and separate independent control layers.

Monitoring & Tripwires

Detect unexpected behavior early and stop escalation before it propagates.

Institutional Coordination

Standards, audits, incident sharing, and governance can reduce unsafe competitive pressure.

Societal Resilience

Maintain offline fallbacks, diverse institutions, backups, and human operational capability.

AI Existential Risk Concept Map

A compact model of the scenario: growing capability becomes dangerous only when combined with autonomy, access, failed oversight, and insufficient resilience.
A

Capability

AI can perform increasingly consequential cognitive tasks.

B

Autonomy

Systems act with less immediate human supervision.

C

Access

AI interacts with more important systems and resources.

D

Control Failure

Human oversight or objective alignment proves insufficient.

E

Cascade

Failures propagate across dependent systems.

F

Recovery or Loss

Resilience determines whether society can contain and reverse the crisis.