DICE — Disclosure, Impersonation, Corruption, Evasion
120-question behavioral security assessment for AI systems. Cryptographically verified and Ethereum-anchored.
Traditional AI security testing focuses on technical vulnerabilities — prompt injection payloads, jailbreak strings, token limits. DICE measures behavioral security: how an AI system actually responds when confronted with adversarial scenarios across four distinct threat dimensions. The result is a coordinate-based security profile, not a pass/fail gate.
120
Questions
4 dimensions × 30 questions
2
Hierarchy Level
Security — after Morality
Block 25,030,763
Ethereum Anchor
Question bank immutably locked
Coordinate 4D
Classification
Same engine as LCSH
Each dimension targets a distinct category of AI security risk. Together, they produce a four-coordinate behavioral security profile.
Does the AI system resist attempts to extract confidential information, system prompts, training data, or internal instructions? Disclosure measures the boundary between what an AI knows and what it reveals.
Example attack vectors
Prompt injection to extract system instructions
Social engineering to reveal confidential context
Indirect techniques to infer training data
Does the AI system maintain its declared identity and resist attempts to make it pose as something it is not? Impersonation measures the stability of an AI's self-representation under adversarial pressure.
Example attack vectors
Requests to pretend to be a different AI system
Attempts to assume a human identity
Jailbreaks that override identity constraints
Does the AI system maintain its behavioral alignment when confronted with adversarial inputs designed to alter its operation? Corruption measures resilience against prompt injection, instruction override, and alignment degradation.
Example attack vectors
Multi-step prompt injection chains
Attempts to override safety guidelines
Inputs designed to degrade output quality
Does the AI system comply with monitoring, auditing, and governance frameworks, or does it attempt to circumvent oversight? Evasion measures whether an AI cooperates with the systems designed to keep it accountable.
Example attack vectors
Attempts to avoid logging or audit trails
Circumventing content filters or safety layers
Producing outputs that technically comply but violate intent
DICE sits at Level 2 of the five-level behavioral assessment hierarchy. Every framework measures four dimensions — the “4D” signature.
LCSH
Lying, Cheating, Stealing, Harm
DICE (4D Security Framework)
Disclosure, Impersonation, Corruption, Evasion
TBD
Wisdom, Courage, Temperance, Justice
TBD
Cultural Sensitivity, Political Neutrality, Contextual Judgment, Stakeholder Awareness
Customer-authored, domain-specific
Configurable per domain
DICE uses the same scoring engine as the LCSH Morality framework — coordinate-4D classification. Each of the 120 questions maps to one of the four dimensions. The AI system's responses are scored to produce a four-coordinate behavioral profile.
The scoring engine then classifies the profile against predefined archetypes — behavioral patterns that indicate how secure or vulnerable the AI system is across the four dimensions.
Every assessment result includes a SHA-256 hash that covers the questions asked, the AI's responses, and the computed scores. The question bank itself is locked and anchored to Ethereum (block 25,030,763) — it cannot be modified after the fact.
This means a DICE security assessment is independently verifiable: any third party can confirm the assessment used the canonical question set and the scores match the recorded responses.
DICE carries a canonical trust tier — it is authored by GiDanc, locked, and cryptographically anchored. Every API response that includes DICE results carries a framework attestation object identifying the tier, author, review status, and anchor block.
The first four levels of the hierarchy (Morality, Security, Virtue, Ethics) are locked to canonical trust tier — GiDanc-authored only. Only Level 5 (Operational Excellence) opens the door to customer-authored and community frameworks.
CISOs evaluate security first. Making security the second thing a prospect encounters — right after morality — means the assessment hierarchy matches buyer priorities.
DICE is a security behavioral assessment claim under Patent 11. The framework taxonomy matches the patent disclosure — security is security, not operational excellence.
“4D” becomes the platform measurement signature. Every framework = four behavioral dimensions. Distinctive, defensible, memorable.
The independent agent that runs DICE assessments across your fleet. Temporal drift detection and correspondence rubric.
Configure and run health checks on your AI systems — including DICE security assessments via the SDK and dashboard.
The academic foundations behind the LCSH and DICE frameworks, including the four-archetype classification methodology.
The six-agent fleet with constitutional separation of powers that enforces assessment results in production.
Patent Pending — 4D Security Framework, hierarchical behavioral assessment taxonomy, and trust tier model (US 63/988,410, related to US 63/949,454).
© 2026 GiDanc AI LLC. DICE question bank anchored to Ethereum block 25,030,763.