Glossary of Terms

A

API Key: Credential for programmatic access to TruthVouch APIs. Scoped, rotatable, and rate-limited.

Alert: Notification triggered by hallucination detection or policy violation. Can be sent via email, Slack, or webhook.

Audit Trail: Immutable, hash-chained log of all requests, decisions, and actions. Tamper-proof.

C

Certification: Cryptographically signed certificate verifying content accuracy (0-100 Trust Score).

Claim: Specific factual assertion extracted from content (e.g., “Founded in 2023”).

Cross-Check: Verification of truth nuggets across multiple AI models simultaneously.

D

Data Subject: Individual whose personal data is processed (GDPR term).

DPA: Data Processing Agreement. Legal contract between Data Controller and Data Processor (GDPR requirement).

Drift: Change in underlying facts after certification, triggering potential certificate revocation.

E

Embedding: High-dimensional vector representation of text meaning. Used for semantic search and caching.

Entailment: Natural Language Inference relationship where premise logically implies hypothesis.

F

Firewall: TruthVouch governance layer that inspects every LLM request/response for policy violations and hallucinations.

Fact Sheet: Structured correction generated automatically when hallucination is detected.

G

GEO: Generative Engine Optimization. Measures content optimization across AI search engines (ChatGPT, Claude, Gemini, etc.).

Governance Pipeline: 15-stage processing system inspecting requests, enforcing policies, and auditing responses.

H

Hallucination: False or unsupported claim made by an LLM. TruthVouch detects with 94%+ accuracy.

HSM: Hardware Security Module. Device storing encryption keys securely, preventing unauthorized access.

I

Injection Attack: Attempt to manipulate LLM through prompt injection. TruthVouch detects and blocks.

M

Multi-Tenancy: Architecture where multiple customers share infrastructure with complete data isolation.

Model Card: Documentation of AI system’s capabilities, limitations, and governance. Required by EU AI Act.

N

Neural Cache: Two-tier caching (Redis L1 + pgvector L2) accelerating fact verification from 100ms to <10ms.

NLI: Natural Language Inference. Measures semantic relationship between statements (entailment/neutral/contradiction).

P

PII: Personally Identifiable Information (email, phone, SSN, etc.). Automatically masked by governance pipeline.

Policy: Governance rule written in Rego language. Controls what LLMs can do, who can use them, etc.

Prompt Injection: Malicious attempt to override LLM instructions via crafted prompts.

R

RAG: Retrieval-Augmented Generation. Technique combining LLMs with knowledge base lookup.

Rego: Policy language (from Open Policy Agent). Used to define TruthVouch governance rules.

S

Shield: Hallucination Shield product. Real-time monitoring across 9+ LLMs.

Sentinel: VT Sentinel Agent. Employee AI tool governance agent for Windows/macOS/Linux.

Sub-Processor: Third party (vendor) who processes data on TruthVouch’s behalf (AWS, Stripe, etc.).

T

Trust Score: 0-100 metric indicating overall content accuracy. Calculated from claim verification.

Truth Nugget: Verified fact in your knowledge base (e.g., company name, founding date, pricing).

TruthOps: Semi-autonomous agents handling monitoring, remediation, compliance tasks.

V

Vendor Assessment: Questionnaire for evaluating AI tools against your requirements (6 dimensions, 3 depth levels).

W

Webhook: HTTP callback notifying your system of TruthVouch events (new alert, certificate revoked, etc.).

Z

Zero Trust: Security model assuming no implicit trust. Every request authenticated and authorized.

---

Need clarification on a term? Email docs@truthvouch.com or start a chat with our support team.