01 · Sensing
Find
A synthetic sensor network detects an object inside a closed training environment.
Daily Break the Chain
Inspect the signal.Change the outcome.
A deterministic, evidence-led systems puzzle. The same synthetic scenario is available to everyone today: inspect what was observable, state your confidence, choose an interruption control, then compare the path not taken.
No registration Synthetic only Desktop + WebXR Local progress
Initializing immersive renderer…
Daily Break the Chain · loading
A chain is forming.
Information is moving between connected nodes. One decision can change the path.
Synthetic scenario · no registration · no real target or external system
System view
Loading the chain…
Select a node to inspect its role without changing the authoritative state.
observed
unverified
human gate
interrupted
Briefing
The Opaque Package
A high-performing model package cannot prove where it came from.
AI supply-chain provenance challenge
Situation
A fictional analytics team receives a model package from an unverified mirror. It passes a basic benchmark and promises better accuracy, but its manifest, build lineage and serialization policy are incomplete.
Your mission
Decide whether the package is merely poorly documented or unsafe to promote, then choose a control that protects the production boundary without blocking approved releases.
Guided mode
Inspect evidence before committing to a hypothesis. High model confidence does not establish source integrity or authority.
Watch the chain form. Select a node in the spatial view or continue when you are ready to inspect the evidence objects.
Evidence objects
0 inspectedOpen the records that matter. Accurate evidence can still be irrelevant.
Commit to an explanation
Choose the causal account best supported by what was observable at the time.
Break the chain
Select the narrowest control that constrains authority while preserving legitimate function.
Immediate consequence
Resolving…
—overall
—assessment
Ground truth
Main insight
Why this outcome occurred
What this result does not mean
This result does not predict real combat outcomes, real infrastructure behavior, or the performance of any actual organization or system.
Counterfactual replay
Compare the path not taken.
Phantom timeline
Alternative loaded
Your path
Alternative
Return without compulsion
Seven deterministic challenges. One local learning record.
Past daily scenarios remain available for review. Progress is stored only in this browser, missed days do not erase earlier work, and no account or behavioral profile is required.
Today
Signal Integrity
AI system security · Orientation
KC-2026-08-01-74E1A4
2026-07-31
The Whisper in the Noise
Cyber defense · Coordinator
KC-2026-07-31-2CCB4E
2026-07-30
Identity Drift
Cyber defense · Analyst
KC-2026-07-30-E4206D
2026-07-29
The Opaque Package
AI system security · Coordinator
KC-2026-07-29-BB5818
2026-07-28
Signal Integrity
AI system security · Orientation
KC-2026-07-28-D7F381
2026-07-27
Operation Glass Aegis
AI system security · Analyst
KC-2026-07-27-51B188
2026-07-26
Identity Drift
Cyber defense · Analyst
KC-2026-07-26-F0124B
0learning days completed—best local assessment
Open analysis sandbox
Trace three different meanings of “kill chain.”
Move freely through military F2T2EA, the traditional cyber intrusion sequence, or an attack chain against an AI system. Enable stage-specific controls and inspect nodes without changing the authoritative stage.
Initializing visual renderer…
Analysis model
Loading simulation…
Preparing the synthetic event stream.
Active
Traversed
Future
Authoritative state
ActiveChain status
0Event sequence
0/0Controls active
0%Progress
Interface-state metrics only—not predictions of operational effectiveness, legality, or harm.
Stage
Initializing
Loading synthetic state.
Observable
Human question
Break the chain
Defensive and governance controls
Enable controls, then advance. A matching control interrupts progression for review; continuing is allowed only for comparative learning.
02 · Identity & location confidence
Fix
Independent synthetic sources are correlated to refine identity and position uncertainty.
03 · Continuity
Track
The synthetic system maintains custody while displaying uncertainty growth during sensor loss.
04 · Decision support
Target
Humans review mission purpose, evidence, alternatives, constraints, and legal requirements.
05 · Authorized simulated effect
Engage
A human-authorized, non-destructive simulated effect is applied inside the training environment.
06 · Effects & accountability
Assess
Post-action evidence is compared with the intended outcome without automatically generating a repeat action.
Guided · Challenge · Analysis
One event model, three ways to learn.
Guided mode explains each decision. Challenge mode adds a visible three-minute clock and rewards calibrated confidence. Analysis mode removes urgency and foregrounds provenance, the event sequence, ground truth, and alternative paths.
Guided
Learn the visual language
Prompts, contextual hints, clear evidence states, and recovery from a poor first decision make the first session understandable without prior training.
Challenge
Commit under uncertainty
A visible timer increases pressure, but speed is only one dimension. Unsupported confidence and disproportionate controls lower the assessment.
Analysis
Freeze, inspect, and compare
Review the append-only event trail, reveal ground truth after resolution, rewind the decision point, and load a phantom alternative without erasing the first path.
Assessment without a destruction score
Measure reasoning, authority, and preservation.
The action report does not count simulated casualties, targets, or damage. It evaluates evidence discipline, confidence calibration, intervention timing, authority containment, functional preservation, and whether the selected control actually interrupted the causal chain.
Evidence discipline
Did the user inspect relevant records, recognize provenance problems, and avoid mistaking accurate but irrelevant data for causal evidence?
Confidence calibration
A Brier-style dimension rewards confidence that matches the correctness of the selected hypothesis and penalizes unsupported certainty.
Authority containment
Did the intervention place a deterministic control at the point where a model, identity, sensor, or process could change external state?
Functional preservation
Could the control stop the unsafe path without unnecessarily disabling trusted monitoring, approved models, legitimate users, or unrelated services?
One phrase, three distinct models
Do not collapse them into one definition.
“Kill chain” can describe a military decision-and-engagement sequence, an intrusion lifecycle, or a staged attack against an AI-enabled application. The site keeps those meanings separate while showing their shared logic: linked dependencies and interruption points.
Military targeting doctrine
Find → Fix → Track → Target → Engage → Assess
AI may assist sensing, fusion, classification, prioritization, guidance, or assessment. That does not by itself establish autonomous authority to use force.
Trace the functional chainCyber intrusion model
Reconnaissance → Actions on Objectives
The classic Cyber Kill Chain presents an ordered campaign narrative. Modern adversary behavior can loop, overlap, or skip stages, so ATT&CK annotations add needed granularity.
Compare cyber stagesAttacks against AI systems
Recon → Poison → Hijack → Persist → Impact
Here the AI system is the target or compromised intermediary. The trust boundary includes models, prompts, retrieval data, memory, tools, identities, and downstream actions.
Explore the AI security chainExpansion horizon
The event model now supports the next immersive layers.
Version 1.2.0 adds the first public Evidence Atlas on top of the deterministic challenge, evidence, scoring, replay, and sharing primitives. Cooperative and authoring systems remain explicitly marked as future releases.
Global Evidence Atlas
Filter country, doctrine, system, human-control, deployment, evidence, timeline, governance, and source layers using country-level abstraction rather than tactical deployment coordinates.
Kill Chains War Room
Two-to-six-person cooperative sessions with asymmetric evidence, role-specific authority, structured recommendations, facilitator controls, and a deterministic team debrief.
Creator Studio
A constrained, declarative scenario editor with synthetic-data enforcement, reviewed components, publishing states, remix lineage, and no arbitrary scripts or external actions.
Trust and publication boundary
Exciting does not mean context-free.
Every challenge is labeled synthetic. Result cards preserve the challenge code and methodological limitation. The friend link hides the sender’s solution until the recipient completes the same deterministic seed. No identity, raw headset pose, gaze, voice, exact location, or private note is included in the challenge state or result card.
The site remains an instructional environment: it does not execute attacks, contact external systems, accept real target data, model casualty outcomes, or provide weapon-design or operational targeting instructions.
Read the evidence and publication methodology, review the privacy design, or open the curated source library.