AI KILL CHAINS & DECISION SYSTEMS
A synthetic crisis simulation showing how automation bias, correlated sensor error, communications delay, preauthorization, and interacting defensive systems can accelerate a local mistake into a wider escalation.
ANALYTICAL & SAFETY BOUNDARIES
Analytical boundary: Autonomy is analyzed by function, context, constraints, evidence, authority, intervention capability, and lifecycle—not as a single label attached to an entire machine. Capability is not deployment; deployment is not autonomous use; recommendation is not authorization; model confidence is not positive identification.
Simulation safety boundary: Synthetic, non-graphic, non-operational education only. Use fictional geography, abstract objects, reserved domains, nonfunctional artifacts, constrained inspect/pause/replay actions, and no real targeting data, attack geometry, payloads, credentials, arbitrary external URLs, malware, command execution, or contact with third-party systems.
Source basis: Exact owner-supplied research-source instances plus bounded official-primary-source currentness v15.
ORIENTATION
Several locally reasonable systems can create a globally dangerous interaction before any operator understands the combined state. The critical problem is not evil machines but architecture, assumptions, interface pressure, and authority inherited from earlier choices.
WORKING BRIEF
The fictional Orison network moves from normal operations to overload, degraded communications, conflicting evidence, and machine-speed interaction. The interface becomes more coercive as timers shrink and caveats collapse.
Multiple feeds may derive from one underlying signal. When fusion treats them as independent, confidence can rise falsely. Automated nodes can then react to one another, turning local defensive rules into a recursive escalation loop.
The ledger records observations, source lineage, model and rule versions, recommendations, human actions, authority changes, communication delay, interventions, external effects, and later corrections. It supports reconstruction but not automatic blame.
COMPLETE DOSSIER
Terms are defined for this site’s evidence method, not as universal legal or clinical definitions.
| Risk | Early warning | Defensive control |
|---|---|---|
| Automation bias | Approval rate rises while evidence quality falls | Independent review prompts and protected hold/reject authority |
| Correlated evidence | Many feeds share one source | Provenance graph and independence score |
| Communications degradation | Approval arrives after action | Predefined safe state and authority expiration |
| Recursive machine response | Systems react to each other | Cross-system damping, rate limits, and human escalation gate |
| Distribution shift | Conditions depart from testing | Outside-assumption indicator and bounded termination |
Pass condition:
Pass condition:
Pass condition:
Pass condition:
Pass condition:
SOURCE QUALITY · UNCERTAINTY · NEUTRALITY
Primary institutional doctrine or policy; supports what the issuing body states, not deployment or compliance.
Primary technical specification or documentation; supports interface/status claims, not truth or field effectiveness.
First-party capability statement requiring independent corroboration.
Attributable public evidence of deployment scope; does not automatically establish autonomy, effectiveness, or legality.
Observed demo or exercise under bounded conditions; not field deployment.
Structured test with stated conditions; external validity remains limited.
Scholarly evidence with method and scope limitations.
Journalistic account requiring attribution and corroboration assessment.
Preserved source packet; claims remain unverified unless separately supported.
Repository-authored inference explicitly marked and linked to supporting evidence.
Synthetic scenario for education; not operational evidence.
Evidence absent from the bounded search; absence is not proof of nonexistence.
Official operator or government statement; supports what that body says, not independent verification of performance.
Independent public evidence supporting a bounded capability, test, status, or deployment proposition.
Material public claim with unresolved attribution, mode, outcome, or corroboration; must remain attributed and nonfinal.
| Meaning | Subject | Analysis model |
|---|---|---|
| AI_AS_TARGET | Attacks against models, data, retrieval, context, tools, infrastructure, and users. | Lifecycle defense: provenance, isolation, least privilege, retrieval authorization, tool-specific credentials, deterministic policy outside the model, egress controls, telemetry, rollback, and human approval. |
| AI_AS_CYBER_ENABLER | AI accelerates conventional reconnaissance, social engineering, vulnerability analysis, or campaign execution. | Defensive analysis must remain non-operational and must not provide executable payloads, credentials, real targets, or attack procedures. |
| AI_ENABLED_MILITARY_KILL_CHAIN | AI assists sensing, fusion, classification, prioritization, assignment, guidance, engagement support, or assessment. | Use a function-context-control model; distinguish recommendation from authorization, navigation autonomy from target-selection autonomy, and classifier score from positive identification. |
Instructional boundary: Explain systems, evidence, uncertainty, failure, oversight, and defense without reproducing targeting software, weapon configuration, attack procedures, evasion methods, or defeat advice.
CLAIM · SOURCE · LIMIT
| Claim block | Evidence class | Source | Supported proposition | Unsupported inference |
|---|---|---|---|---|
| five-rounds | OWNER_SUPPLIED_RESEARCH_SYNTHESIS | The Last Human Decision Speed, Automation Bias, Failure, and Accountability.md | The fictional Orison network moves from normal operations to overload, degraded communications, conflicting evidence, and machine-speed interaction. The interface becomes more coercive as timers shrink and caveats collapse. | Does not by itself establish deployment, exact operating mode, combat use, effectiveness, legality, consensus, or endorsement. |
| correlated-error | OWNER_SUPPLIED_RESEARCH_SYNTHESIS | Algorithmic Gatekeeper Interactive Specification.md | Multiple feeds may derive from one underlying signal. When fusion treats them as independent, confidence can rise falsely. Automated nodes can then react to one another, turning local defensive rules into a recursive escalation loop. | Does not by itself establish deployment, exact operating mode, combat use, effectiveness, legality, consensus, or endorsement. |
RESEARCH EDITION
This page follows the public method for provenance, confidence, source independence, alternative accounts, limitations, review state, and visible correction.
CONTINUE