Core Concepts
- Shadow AI
- The use of artificial intelligence tools, platforms, or AI-powered workflows at work without formal approval, oversight, or governance from the organization’s IT, security, compliance, or legal teams. Shadow AI is a subset of Shadow IT specific to AI tools. Usage is typically not malicious — employees adopt AI tools for productivity reasons — but it creates data exposure, compliance risk, and audit trail gaps the organization cannot manage because it does not know the usage is occurring. See What Is Shadow AI? for the full definition and context.
- Shadow IT
- The use of any technology — software, cloud services, devices, applications — at work without formal IT approval, procurement, or oversight. Shadow IT has been documented in enterprise environments since the cloud era of the 2000s–2010s. Shadow AI is a specific subset of Shadow IT focused on AI tools. The key difference: AI tools require no installation and are freely accessible via browser, making Shadow AI far harder to detect and control than earlier categories of Shadow IT. See Shadow AI vs Shadow IT.
- Sanctioned AI
- An AI tool or AI-powered workflow that has been formally reviewed, approved, and authorized by the organization’s IT, security, legal, and/or compliance teams. Sanctioned AI operates under a vendor agreement (typically including a data processing agreement or DPA), within defined use-case boundaries, and with organizational visibility into what data is processed. The goal of Shadow AI governance is to migrate unsanctioned AI use to sanctioned alternatives rather than simply prohibiting AI use.
- Generative AI (GenAI)
- A category of AI systems that generate new content—text, images, code, audio, or video—in response to prompts, rather than performing classification or prediction tasks on existing data. Large language models (LLMs) such as GPT-4, Claude, and Gemini are the most widely used generative AI systems in workplace contexts. Generative AI is the primary driver of the current Shadow AI problem: these tools are free, browser-accessible, and capable enough to genuinely accelerate work tasks, making employee adoption extremely rapid.
- Large Language Model (LLM)
- A type of generative AI model trained on large volumes of text data to predict and generate human-like text. LLMs are the underlying technology in ChatGPT, Claude, Gemini, and most consumer and enterprise AI writing and coding assistants. When employees paste company data into an LLM-based chatbot, that data is submitted to the model’s provider under that provider’s terms of service—which may include rights to use submitted data for model training unless an enterprise or privacy agreement prohibits it.
- AI Governance
- The organizational policies, processes, roles, controls, and accountability structures that ensure AI systems are developed, deployed, and used in a manner consistent with the organization’s risk tolerance, legal obligations, and ethical standards. Effective AI governance programs typically cover AI inventory management, risk assessment, policy enforcement, employee training, vendor management, and incident response. Shadow AI—AI use outside any governance structure—represents a direct gap in AI governance coverage.
- AI Acceptable Use Policy
- A written policy defining which AI tools employees may use, under what conditions, and with what data types. The most effective AI policies define an approved tools list, a data classification framework for AI use, a tool approval request process, and a graduated enforcement approach. See Shadow AI Policy for a full guide and framework template.
Risk and Compliance Terms
- Data Exposure
- The transmission of sensitive, confidential, or regulated data to a system or party outside the organization’s controlled environment without authorization or appropriate safeguards. In the Shadow AI context, data exposure occurs when an employee submits customer records, source code, financial data, health information, or other sensitive content to a public AI tool whose data handling terms do not meet the organization’s obligations.
- Data Processing Agreement (DPA)
- A contractual agreement between a data controller (the organization) and a data processor (the AI vendor) specifying how personal data will be handled, protected, and processed. Required under GDPR (Article 28) and UK GDPR for any vendor that processes personal data on behalf of the controller. Most consumer AI tools do not offer DPAs for free-tier accounts—meaning personal data submitted to these tools is processed without the contractual protections GDPR requires. Enterprise AI agreements typically include DPAs.
- Business Associate Agreement (BAA)
- A contract required under HIPAA between a covered entity (healthcare provider, health plan, or healthcare clearinghouse) and any business associate (vendor, contractor, or service provider) that creates, receives, maintains, or transmits Protected Health Information (PHI) on their behalf. Without a signed BAA, submitting PHI to an AI tool is a HIPAA violation. Most public AI tools do not offer BAAs for consumer or free-tier accounts; enterprise healthcare AI agreements typically do.
- Protected Health Information (PHI)
- Under HIPAA, any individually identifiable health information held or transmitted by a covered entity or its business associates—including patient names, dates of service, diagnoses, treatment records, insurance information, and other health-related data. PHI is one of the most restricted data categories in the Shadow AI context: it must not be submitted to any AI tool without a signed BAA and an assessment of HIPAA compliance.
- Personally Identifiable Information (PII)
- Information that can be used to identify a specific individual, either directly (name, Social Security number, email address) or in combination with other data. PII is regulated under GDPR, CCPA, and numerous other privacy laws. Submitting PII to a public AI tool without a DPA and a lawful basis assessment creates regulatory exposure under applicable data protection law.
- Hallucination (AI)
- A term used in AI safety and evaluation contexts to describe AI-generated output that is plausible and fluently presented but factually incorrect, fabricated, or unsupported by the input. AI hallucinations are a documented risk in legal research (fabricated case citations), medical summarization (incorrect drug information), and financial analysis (incorrect figures presented as facts). Organizations relying on AI outputs without a required human review step are exposed to hallucination-driven errors in high-stakes decisions. Also referred to in formal contexts as “confabulation” (the term used in NIST AI 600-1).
- Audit Trail
- A chronological record of system activity that is sufficient to reconstruct and examine the sequence of environments and activities surrounding events. In the Shadow AI context, audit trail gaps occur when employees use personal or unapproved AI accounts for work tasks—these accounts generate no organizational record, meaning a breach investigation, compliance audit, or litigation disclosure cannot establish what data was submitted, when, or by whom. The absence of an audit trail is one of the six primary Shadow AI risk categories.
Regulatory Terms
- GDPR (General Data Protection Regulation)
- Regulation (EU) 2016/679, the EU’s comprehensive data protection law, in force since May 2018. Applies to the processing of personal data of EU residents, regardless of where the processing organization is located. Key principles relevant to Shadow AI include lawful basis for processing, data minimization, purpose limitation, storage limitation, and security. Shadow AI creates GDPR exposure when employees submit personal data of EU residents to AI tools without a lawful basis, a DPA, and appropriate safeguards. Maximum penalties: up to €20M or 4% of global annual turnover.
- HIPAA (Health Insurance Portability and Accountability Act)
- US federal law governing the protection of Protected Health Information (PHI). The HIPAA Privacy Rule restricts who may access PHI and for what purposes; the Security Rule requires administrative, physical, and technical safeguards for electronic PHI. Shadow AI creates HIPAA exposure when healthcare employees submit PHI to AI tools without a signed BAA. See Shadow AI in Healthcare.
- EU AI Act
- Regulation (EU) 2024/1689, the EU’s risk-tiered AI regulation, entered into force August 1, 2024. Classifies AI into four risk tiers: prohibited (Article 5), high-risk (Annex III), transparency-obligated, and general-purpose AI. Directly relevant to shadow AI through Article 26 (deployer obligations for high-risk AI, including worker consultation requirements) and Article 5(1)(f) (prohibition on AI that infers emotions in workplace settings). See International AI Regulations.
- Deployer (EU AI Act)
- Under the EU AI Act (Article 3(4)), a deployer is “a natural or legal person, public authority, agency or other body using an AI system under its authority.” Deployers of high-risk AI systems have obligations under Article 26, including human oversight requirements, incident reporting, and worker consultation before deploying AI systems that affect employees. Shadow AI creates a novel deployer-obligation problem: if an employee uses a high-risk AI tool without organizational knowledge, the organization may have incurred Article 26 deployer obligations it cannot fulfill because it does not know the system is in use.
- SOX (Sarbanes-Oxley Act)
- US federal law (2002) governing financial reporting requirements for public companies. Sections 302 and 404 require documented internal controls over financial reporting, with executive certification and auditor attestation. Shadow AI creates SOX exposure when employees use unapproved AI tools to prepare, analyze, or summarize financial data that feeds into SEC filings or financial statements, because no AI-specific internal control has been assessed, documented, or tested.
Governance Framework Terms
- NIST AI RMF (AI Risk Management Framework)
- Published by NIST as AI 100-1 on January 26, 2023. A voluntary, non-certifiable framework organizing AI risk management into four core functions: Govern (cross-cutting policy and culture), Map (AI inventory and risk identification), Measure (risk assessment), and Manage (risk treatment and monitoring). Shadow AI directly undermines the Map function: AI systems in use without organizational knowledge cannot be inventoried or assessed. See Governance Frameworks.
- ISO/IEC 42001:2023
- The first certifiable international standard for AI management systems (AIMS), published December 2023. Follows a Plan-Do-Check-Act structure with an Annex A control library. Unlike the NIST AI RMF, ISO/IEC 42001 supports third-party certification, making it the standard of choice for organizations that need to demonstrate AI governance maturity to external stakeholders. Shadow AI creates a gap in ISO/IEC 42001 scope: unsanctioned AI use falls outside the management system boundary.
- AIMS (AI Management System)
- A system of policies, processes, and controls for managing AI risks and opportunities within an organization, structured according to the Plan-Do-Check-Act model. ISO/IEC 42001 is the primary certifiable AIMS standard. An AIMS addresses the full AI lifecycle: procurement, development, deployment, monitoring, and decommissioning of AI systems within scope.
- Govern (NIST AI RMF)
- The cross-cutting core function of the NIST AI RMF. Govern establishes the organizational culture, accountability structures, policies, and processes that make the Map, Measure, and Manage functions possible. In practice, Govern includes AI strategy, policy development, role assignment, risk tolerance definition, and employee training. Effective Shadow AI governance requires a strong Govern function: the policy, culture, and accountability structures that make employees willing to use sanctioned tools and report incidents.
AI System Types
- Consumer AI Tool
- An AI tool offered directly to individual users, typically via a web browser or mobile app, under terms of service designed for individual rather than enterprise use. Consumer AI tools are the primary source of Shadow AI risk: they are free, instantly accessible, require no IT involvement, and are often used by employees for work tasks without awareness of the data-handling implications. Consumer accounts typically do not include enterprise data protection terms (DPAs, BAAs, data isolation).
- Enterprise AI
- AI tools and platforms sold to organizations under enterprise agreements that include data processing agreements, data isolation commitments, access controls, audit logging, and other security and compliance features absent from consumer tiers. Enterprise AI is the sanctioned alternative to consumer AI for most workplace use cases. Common examples include Microsoft 365 Copilot, ChatGPT Enterprise, Google Workspace AI features, and enterprise-tier coding assistants. Enterprise agreements typically include BAA availability for healthcare customers and GDPR DPA terms for EU operations.
- Agentic AI
- AI systems that autonomously plan, execute multi-step tasks, and interact with external tools, APIs, or data sources without requiring a human to approve each action. Agentic AI creates elevated Shadow AI risk compared to chatbots: a sanctioned agentic AI workflow operates within defined data access boundaries; an unsanctioned one may autonomously access, process, or transmit organizational data in ways that are difficult to reconstruct after the fact. Agentic AI governance is an emerging area in enterprise AI policy.
- AI Browser Extension
- A browser plugin or extension that adds AI-powered capabilities to the user’s browser environment, often including page summarization, email drafting, writing assistance, or research tools. AI browser extensions are a significant Shadow AI vector because they can access page content (including web-based applications like CRMs, email, and project management tools) and transmit that content to the extension’s AI backend. Many employees install AI browser extensions without realizing the data access implications. Policy scope should explicitly include AI browser extensions.
Free Resource
Shadow AI Assessment Checklist
A practical checklist for evaluating your organization's Shadow AI exposure across discovery, policy, controls, training, and compliance. Download and use it as a starting point for your governance review.
Frequently Asked Questions
What is the difference between Shadow AI and unsanctioned AI?
The terms are used interchangeably in most industry and research contexts. 'Shadow AI' emphasizes the hidden/invisible nature of the usage — AI operating outside organizational visibility, like a shadow. 'Unsanctioned AI' is a more neutral descriptor focusing on the absence of formal authorization. Both refer to the same phenomenon: AI tool use at work without IT/security/compliance approval.
Is 'confabulation' the same as 'hallucination' in AI?
In practice, yes — both terms describe AI-generated content that is factually incorrect, fabricated, or unsupported by the input. 'Hallucination' is the more common industry term. 'Confabulation' is the term used in NIST AI 600-1 (the GenAI Profile), borrowed from psychology where it describes a memory error involving fabricated but sincerely believed information. The NIST usage reflects a preference for technically precise terminology in formal standards contexts.
Cite This Page
APA-style
Shadow AI Guide. (2026). Shadow AI Glossary: Key Terms and Definitions. Retrieved from https://www.shadowaiguide.com/shadow-ai-glossary