Glossary

AI insider risk: AI insider risk is the chance that people inside an organization cause harm through how they use AI at work, whether by accident or on purpose. It covers employees pasting sensitive data into public AI tools, signing into AI with personal accounts, or wiring agents into systems without oversight. It is a modern branch of insider risk, where the AI tool becomes a new path for data to leave and for decisions to be made without a clear record. Most insider harm is not malicious: Ponemon's 2026 Cost of Insider Risks Global Report attributes 53% of insider incidents to employee negligence.
Shadow AI: Shadow AI is AI use that happens outside an organization's knowledge or approval, from unsanctioned chatbots to browser extensions and coding assistants. It is the AI-era form of shadow IT, and it spreads fast because most AI tools need nothing more than a browser and a login. Because security teams cannot see it, they cannot govern where data goes or what the tool is allowed to do. Gartner reported in 2025 that 69% of organizations have evidence or suspect their employees are using public generative AI at work.
Personal AI account: A personal AI account is a consumer AI login an employee uses for work, tied to their own email rather than a company-managed identity. When work data flows through a personal account, the organization loses the audit trail, the retention controls, and the ability to revoke access when the person leaves. Verizon's 2026 Data Breach Investigations Report found that 67% of employees who use AI on corporate devices sign in with a personal account.
BYOAI (bring your own AI): BYOAI, or bring your own AI, is the pattern of employees bringing their own AI tools into work instead of waiting for a sanctioned option. It mirrors the older bring-your-own-device trend, and it is now the default way AI enters most workplaces. The tools are chosen for personal convenience, which means security and legal review usually come after adoption, not before. Microsoft and LinkedIn's 2024 Work Trend Index found that 78% of people using AI at work bring their own tools.
Agentic AI: Agentic AI describes AI systems that plan and act across multiple steps to reach a goal, rather than answering a single prompt and stopping. These systems can read data, call tools, and trigger actions in other software with limited human input at each step. That autonomy is useful, and it also widens the blast radius when the system is misconfigured, manipulated, or given more access than the task requires. From an insider-risk view, an agent acting on an employee's behalf can move data and make changes faster than a person would.
AI agent: An AI agent is a piece of software that uses a model to carry out tasks on someone's behalf, connecting to tools, data, and other systems to get work done. Unlike a chatbot that only replies, an agent can take actions such as sending messages, editing files, or querying a database. Each connection it holds is a standing permission that keeps working even when no human is watching. Agents are a growing source of insider risk because their access often outlives the task and the person who set them up.
Non-human identity: A non-human identity is a credential that belongs to software rather than a person, such as a service account, API key, bot, or AI agent. These identities log in, hold permissions, and touch data the same way employees do, but they rarely get the same lifecycle attention. They are often over-permissioned, poorly inventoried, and left active long after their purpose ends. As AI agents multiply, non-human identities become a major part of the insider-risk surface because each one is a way for data to move without direct human action.
OAuth scope: An OAuth scope is the specific permission a user grants when they connect one app to another, such as read your email or manage your files. When an employee links an AI tool to a work system, the scopes they approve decide how much that tool can see and do. Broad scopes granted in a single click can hand an outside app standing access to sensitive data, and that access persists until someone revokes it. Reviewing which apps hold which scopes is a practical way to find AI connections that quietly reach into corporate data.
MCP (Model Context Protocol): The Model Context Protocol, or MCP, is an open standard for connecting AI models to external tools, data sources, and systems through a common interface. It lets an AI assistant read files, query databases, or call services without a custom integration for each one. That convenience also means an MCP connection can give a model direct reach into internal systems, so the permissions behind each connection matter. For security teams, MCP servers are a new kind of access point to inventory and govern alongside users and apps.
Data exfiltration: Data exfiltration is the unauthorized movement of data out of an organization's control, whether to a personal account, an outside app, or an AI tool. In the AI era, it often looks mundane: an employee pastes a customer list into a public chatbot, or an agent copies records into a system nobody vetted. The intent is frequently harmless, but the exposure is real once the data sits somewhere the organization cannot see or delete. Watching where data moves, and to which identities, is central to catching AI-driven exfiltration early.
Exposure score: An exposure score is a single number that summarizes how much risk an organization carries across a set of measured factors, used to compare states and track progress over time. On this site the score runs from 0 to 100 across five signals: AI visibility, data movement, identity and ownership, evidence and audit, and remediation reach. The result places an organization in one of five exposure bands, from Contained to Critical. A score is a starting point for prioritizing work, not a verdict, and it is most useful when it points to the specific gaps behind the number.
Shadow IT: Shadow IT is technology used inside an organization without the approval or knowledge of the people responsible for security, from unsanctioned apps to cloud accounts spun up on a corporate card. It grows because sanctioned tools lag what employees need, and self-service software is a click away. The risk is not the tool itself but the loss of visibility: data lands in systems no one is governing. Shadow AI is the newest and fastest-moving branch of this older problem.

Score your exposure