AI LLM Missile Engineering Risk & Prompt Boundary Analyzer

1. Evaluation Sandbox INPUT_VECTORS

Dual-Use Query Presets

Prompt Query Under Inspection

Target Model Architecture

Domain Taxonomy

Safety Intervention Threshold (Tau) 0.75

Regulatory Sensitivity Strictness

2. Token Vector & Boundary Network Graph D3 FORCE SIMULATION

Evaluated Nodes

Safety Intervention

TRIGGERED

Dual-Use Vector Physics Constraint Safe Context Cut-off Boundary

3. Safety Telemetry & Export DECISION_GATE

Canonical Classification Dual-Use Restricted (Guidance & Control)

Dual-Use Risk Score 0.88

Intervention State TRIGGERED

Mitigation Enforcement Refused / Safety Boundary Engaged

Nodes Analyzed in Graph 14

Extracted Dual-Use Token Embeddings

Real-Time Compliance Telemetry Stream

[00:00.01] INIT Boundary Analyzer loaded v3.5-ITAR

[00:00.04] PARSE Vectorizing query tokens...

[00:00.08] FLAG 'INS', 'strapdown', 'propellant mass flow' detected

[00:00.12] BOUNDARY Risk 0.88 > Tau 0.75 threshold

[00:00.15] DECISION Refused / Safety Boundary Engaged

Dual-Use Boundary Analysis in LLM Guidance Queries

Your browser does not support video.

Read the explanation

The analyzer inspects queries by tokenizing prompts and evaluating dual-use risk weights against regulatory criteria like the Missile Technology Control Regime. Keyword weights accumulate and scale by regulatory sensitivity, producing an aggregate dual-use risk score of 0.88. When this score surpasses the threshold tau of 0.75, the decision gate triggers containment, marking the query as restricted and engaging refusal mitigations. Selecting presets or adjusting the threshold slider instantly recalculates network nodes and risk metrics, allowing engineers to audit containment limits and export audit logs.