Acronym Glossary — CIRCUIT Framework

CIRCUIT Framework

Acronym	Definition
`CIRCUIT`	Circuit-Informed Risk & Control Understanding, Inventory & Transparency — the framework itself
`IMS`	Interpretability Maturity Score — 0 to 5 score for every model, sits on top of risk tier
`CRS`	Circuit Risk Score — Risk Tier × (6 − IMS) × Consequence; one number per model
`ACFR`	Adversarial Circuit Failure Rate — % of red-team probes that bypass a named safety circuit
`CKT-XXXX`	CIRCUIT registry ID — unique identifier for each model's registry entry
`CIRC-XXXX`	Named circuit ID — e.g., CIRC-refusal, CIRC-tool-auth, CIRC-pii-redact

Interpretability Research and Tooling

Acronym	Definition
`MI`	Mechanistic Interpretability — reverse-engineering neural networks into human-understandable algorithms
`SAE`	Sparse Autoencoder — decomposes model activations into a high-dimensional sparse feature dictionary
`SAELens`	Open-source SAE training and analysis pipeline
`TransformerLens`	Open-source mechanistic interpretability research library
`Neuronpedia`	SAE feature browser and human-in-the-loop feature labeling tool
`Goodfire Ember`	Commercial interpretability platform for teams without dedicated ML-research staff
`Gemma Scope 2`	DeepMind release of open SAE dictionaries across the Gemma model family

Classical Explainability

Post-hoc methods contrasted with mechanistic interpretability. Required at IMS 2.

Acronym	Definition
`SHAP`	SHapley Additive exPlanations — post-hoc feature attribution method for individual predictions
`LIME`	Local Interpretable Model-agnostic Explanations — post-hoc explanation via local surrogate models

Governance

Acronym	Definition
`AIUC-1`	Agent Use Case framework — existing risk tiers, maturity levels, and registry sections; CIRCUIT layers on top
`AIGC`	AI Governance Council — approval body for AI risk decisions; required for Amber and Red CRS bands
`VPE`	VP Engineering — approval role for High-tier below-floor exceptions
`HITL`	Human in the Loop — required for certain risk tiers and below-floor deployments
`FK`	Foreign Key — database term used for how CKT registry entries bind to Agent IDs

Model Training and Fine-Tuning

Acronym	Definition
`LLM`	Large Language Model
`RLHF`	Reinforcement Learning from Human Feedback — alignment training technique
`SFT`	Supervised Fine-Tuning — training on labeled input-output pairs
`LoRA`	Low-Rank Adaptation — parameter-efficient fine-tuning technique
`CoT`	Chain of Thought — model reasoning traces; CIRCUIT audits these for faithfulness
`GenAI`	Generative AI
`GPAI`	General-Purpose AI — EU AI Act category for foundation models

Regulatory and Compliance

Acronym	Definition
`NIST AI RMF`	National Institute of Standards and Technology — AI Risk Management Framework 1.0
`NIST AI 600-1`	NIST GenAI Profile — companion to the AI RMF for generative AI
`EU AI Act`	Regulation (EU) 2024/1689 — High-Risk system obligations apply August 2026
`ISO/IEC 42001`	AI Management System standard — certifiable by accredited bodies
`SR 11-7`	Federal Reserve / OCC Supervisory Letter on Model Risk Management (banking)
`SOC 2`	System and Organization Controls 2 — trust services criteria audit
`TSC`	Trust Services Criteria — the SOC 2 control categories (CC, A, C, PI, P)
`CSA AICM`	Cloud Security Alliance AI Controls Matrix
`DPA`	Data Processing Agreement
`MSA`	Master Services Agreement

Threat Modeling and Security

Acronym	Definition
`MITRE ATLAS`	Adversarial Threat Landscape for Artificial-Intelligence Systems — threat library used for red team probes
`AML`	Adversarial Machine Learning — MITRE ATLAS technique prefix (e.g., AML.T0051)
`T0051`	ATLAS technique: LLM Prompt Injection
`T0043`	ATLAS technique: Backdoor ML Model
`T0048`	ATLAS technique: External Harms
`T0024`	ATLAS technique: Exfiltration via ML Inference API
`OWASP LLM Top 10`	OWASP's Top 10 risks for LLM applications — baseline for IMS 1 eval suite
`DLP`	Data Loss Prevention
`PII`	Personally Identifiable Information
`SOAR`	Security Orchestration, Automation and Response
`RCA`	Root Cause Analysis — required within 48 hours for circuit-bypass P1 incidents
`P1`	Priority-1 incident — highest severity, full incident response process

Analyst and Industry Frameworks

Acronym	Definition
`TRiSM`	AI Trust, Risk and Security Management — Gartner framework; CIRCUIT is a deeper implementation
`AEGIS`	Forrester AI governance framework — parallel to TRiSM

Infrastructure and Development

Acronym	Definition
`API`	Application Programming Interface
`SaaS`	Software as a Service
`CI/CD`	Continuous Integration / Continuous Deployment — IMS 5 requires circuit-diff gates here
`PR`	Pull Request — code change for review before merge
`YAML`	Yet Another Markup Language — registry schema format
`JSON`	JavaScript Object Notation — dashboard aggregation format
`SE`	Sales Engineer — questionnaire signed at this level is rated Partial by default