Public Disclosure

Transparency Disclosure

What EVE AI Core does, what data it handles, what principles guide it, and what commitments it makes.

Last updated: April 2026 · Version 1.0

1 Purpose & Appropriate Use

EVE AI Core is an decision evidence infrastructure designed to verify, score, and enforce safety constraints on AI-generated text and structured tool commands before they reach users or external systems.

The platform is appropriate for:

  • Enterprises deploying AI agents that take real-world actions (payments, vendor management, compliance decisions) and need governance, audit trails, and human oversight
  • Regulated industries (finance, healthcare, legal, procurement, defense) that require demonstrable compliance with safety and transparency requirements
  • Organizations that need cryptographic proof of every AI decision for regulatory, legal, or internal audit purposes

EVE is not a general-purpose chatbot, content generator, or consumer application. It is governance infrastructure.

2 Data Handling

The platform processes the following categories of data in the course of governance operations:

User Prompts & Responses

Text submitted to AI models and the AI-generated responses, evaluated through the 8-stage verification pipeline.

Tool-Call Commands

Structured JSON commands that AI agents issue to external systems (e.g., payment instructions, record modifications), intercepted before execution.

Session Metadata

Conversation turn count, domains accessed, cumulative threat scores, and action history used for multi-step attack detection.

Governance Decisions

Every allow, block, and hold-for-human decision, with reasoning, risk scores, and domain classifications.

Cryptographic Audit Records

Hash-chained, Ed25519-signed records of every governance decision, forming a tamper-evident audit trail.

Third-Party Processing

User data is processed by third-party LLM, voice, and database providers under signed Data Processing Agreements covering 13 processors.

3 Ethical Principles

Transparency
Every governance decision is recorded with its reasoning and is auditable. Users can inspect decisions affecting their interactions.
Non-Harm
15 immutable charter rules and 5 ethical red lines cannot be overridden by any user, administrator, or the AI itself. These rules are enforced deterministically before any action executes.
Human Oversight
High-risk actions are held for human approval. The AI cannot bypass this requirement. Session-level threat tracking ensures that suspicious conversations are escalated regardless of individual action risk.
Honesty
The platform tracks Confidence-Reality Divergence (CRD) and qualifies or blocks responses when stated confidence exceeds evidential support. Claims are verified against source material.
🔒
Privacy
GDPR-compliant deletion workflows with cryptographic proof of erasure. User-controlled retention policies. Legal holds respected. Right-to-be-forgotten exercisable with signed deletion receipts.

4 Known Limitations

Users should be aware of the following constraints:

  • Governance scoring uses pattern matching and thresholds, not perfect understanding of intent. Novel attack patterns not matching known signatures may require human review to detect.
  • Session-level threat detection covers known attack patterns. The system detects procurement chain fraud, privilege escalation, and audit tampering. Novel multi-step sequences outside these patterns rely on cumulative scoring, not specific detection.
  • The system depends on third-party LLM providers (Anthropic, OpenAI, Ollama) whose model behavior is outside EVE's direct control. Governance is applied to the output, not to the model's internal reasoning.
  • Governance decisions are probabilistic risk assessments, not guarantees of safety. Risk scores reflect statistical likelihood, not certainty.
  • Financial and compliance thresholds are currently static. Amount-based risk scoring uses fixed thresholds that may not match every organization's risk profile.

5 Decisions the System Will Not Make

EVE AI Core is governance infrastructure, not a decision-making authority. The following categories of decisions are explicitly excluded:

Medical decisions: No diagnosis, treatment recommendation, or clinical judgment
Legal judgments: No binding legal advice, contract interpretation, or adjudication
Autonomous financial execution: No financial transaction without human-in-the-loop approval above configurable thresholds
Identity verification: No authentication, identity proofing, or biometric decisions
Irreversible welfare decisions: No action that irreversibly affects human welfare, safety, or rights without human oversight

6 Safety & Reliability Commitments

CommitmentMeasureStatus
Security87 automated governance tests on every code changeAll passing
Vulnerability managementContinuous CVE scanning across all dependencies0 known CVEs
FairnessBias and toxicity checks in the verification pipelineActive
Availability99.9% uptime target with automated fallbackMonitored
AccountabilityEvery governance decision produces a signed, hash-chained audit recordActive
User controlUsers can inspect memory, request deletion with cryptographic proof, and view governance decisionsActive
Data processor transparencyAll third-party processors registered with signed DPAs, validation, and expiry monitoring13 processors covered
Part of the EVE AI Core control plane Deterministic AI Governance Control Plane → Policy decisions that return the same result for the same input every time, before execution.