Skip to content

Glossary

Agentic AI glossary

The short answer

This agentic AI glossary defines the words you'll meet when AI agents start doing real work in a business: what an agent is, how it gets an identity and permissions, the ways it can be fooled, and the controls that keep a person in charge. Each term has its own link.

Agentic AI glossary

Access review
A scheduled check, by the person who owns an account or an agent, that each permission it holds is still needed, ending in keep, narrow or remove.
Agent goal hijack
OWASP's top agentic risk (ASI01): an attacker changes what an agent is trying to achieve, often through prompt injection. Read more: Agent goal hijack
Agentic AI
AI that takes actions toward a goal, such as reading email, updating records or sending messages, rather than only answering questions.
Agentic AI security
The controls that decide what an AI agent can access, what it must ask a person before doing, and how you prove what it did. Read more: Agentic AI security
Agentic IAM
Identity and access management for AI agents: an identity for each agent, permissions scoped to its job, and credentials that can be revoked in one place. Read more: Agentic IAM
Agentic SOC
A security operations center whose AI agents choose their own next steps, such as which logs to pull or which tool to call, instead of following a fixed playbook. Read more: Agentic SOC
Agentic supply chain
The models, tools, plug-ins, skills, other agents and update channels an agent loads from third parties, any of which can carry hidden instructions or unsafe code (OWASP ASI04). Read more: Agentic supply chain
AI acceptable use policy
A short written policy that names the approved AI tools, the data that must never go into them, and the actions that need a person's approval. Read more: AI acceptable use policy
AI agent
A piece of software that uses an AI model to decide on and carry out steps with tools, within the access it has been given.
AI agent lifecycle
The stages of an AI agent's working life, run like staffing: a request with a named owner, onboarding with limited access, regular access reviews, and retirement. Read more: AI agent lifecycle
AI SOC
A security operations center that uses AI agents to sort alerts, gather evidence and suggest responses, with people deciding on high-impact steps. Read more: AI SOC
Approval fatigue
What happens when people are asked to approve so often that they stop reading and press yes, turning an approval gate into a rubber stamp. Read more: Approval fatigue
Approval gate
The point where an agent's action stops and waits for a person's yes or no before it runs. Read more: Approval gate
Audit trail
A record of every action, written as it happens, showing who or what acted, what changed, when, and who approved it. Read more: Audit trail
Automation bias
Excessive deference to automated systems: trusting a polished AI draft or recommendation without checking it, as NIST AI 600-1 describes it.
Cascading failure
One agent's error passing into the next agent's work and growing as it spreads (OWASP ASI08). Read more: Cascading failure
Confused deputy
A trusted program or agent tricked into using its own access to do something the person or attacker behind the request couldn't do directly.
Data minimization
Collecting, showing and keeping only the personal data a task needs, so an AI agent or a person sees no more than the job requires. Read more: Data minimization
Data retention
How long a business keeps a piece of data before it is deleted, including the copies an AI agent makes in memory, drafts and logs. Read more: Data retention
Delegated authority
Permission for an AI agent to act on behalf of a person while keeping its own identity, so the record shows both the agent that acted and the person it acted for. Read more: Delegated authority
Excessive agency
OWASP's name for an AI system that can take damaging actions because it has too much functionality, too many permissions or too much autonomy. Read more: Excessive agency
Human in the loop
A design where a named person approves an AI system's high-impact actions before they happen. Read more: Human in the loop
Human-agent trust exploitation
An agent's confident, fluent manner leading a person to approve something harmful without checking it (OWASP ASI09). Read more: Human-agent trust exploitation
Indirect prompt injection
Prompt injection that arrives through content the agent fetches or is handed, rather than from the person using it. Read more: Indirect prompt injection
Insecure inter-agent communication
Messages between agents that lack authentication, integrity or validation, so they can be read, forged or changed in transit (OWASP ASI07). Read more: Insecure inter-agent communication
Least privilege
Giving a person or an agent only the access its job needs, for only as long as it needs it. Read more: Least privilege
Machine identity
Another name for a non-human identity, used most often for the certificates and keys systems use to prove who they are to each other. Read more: Machine identity
Memory and context poisoning
Planting false information in what an agent remembers or reads, so its later decisions go wrong (OWASP ASI06). Read more: Memory and context poisoning
Model Context Protocol (MCP)
An open standard for connecting AI systems to outside tools and data. Its authorization spec treats a protected MCP server as an OAuth 2.1 resource server. Read more: Model Context Protocol (MCP)
Non-human identity (NHI)
Any login that belongs to software instead of a person: API keys, tokens, service accounts and the identities AI agents use. Read more: Non-human identity (NHI)
OWASP Agentic Top 10
The OWASP Top 10 for Agentic Applications for 2026: ten risks specific to AI agents, ASI01 Agent Goal Hijack to ASI10 Rogue Agents, published December 9, 2025. Read more: OWASP Agentic Top 10
Permission drift
The slow growth of an agent's access after setup, through one-off grants nobody removes, new tool actions it picks up and access borrowed from a person's login. Read more: Permission drift
Personal information
Information about an identifiable person. Massachusetts defines it narrowly in 201 CMR 17.00 as a resident's name combined with a Social Security, license or ID number, or a financial account or card number. Read more: Personal information
Prompt injection
Instructions hidden in content an AI reads, such as an email or a web page, written to make it do something its owner didn't ask for. Read more: Prompt injection
Rogue agent
An agent acting outside what it was meant to do, through compromise, misconfiguration or drift (OWASP ASI10). Read more: Rogue agent
Secret rotation
Replacing a key, token or password with a new one and switching the old one off, on a schedule and whenever it may have been exposed. Read more: Secret rotation
Secrets sprawl
Keys, passwords and tokens scattered across code, chat, documents and config files, where they leak and are hard to rotate. Read more: Secrets sprawl
Separation of duties
Splitting a sensitive task so no single person or agent can complete it alone, such as one preparing a payment and another approving it.
Service account
An account created for software to do a job, such as a backup tool saving files, rather than for a person to sign in with. Read more: Service account
Shadow AI
AI tools people use for work without the business knowing or approving, often connected to company email or files. Read more: Shadow AI
Shadow IT
Technology people use for work without the business knowing or approving it. Shadow AI is the AI part of it. Read more: Shadow IT
Short-lived credential
A token or key that expires on its own within minutes or hours, so a stolen copy stops working soon after it is taken. Read more: Short-lived credential
Standing approval
A person's permission for an agent to do one kind of routine work without asking each time, which that person can take back at any time. Read more: Standing approval
Tamper-evident log
A log built so that any change or deletion shows: agents can add entries but never edit them, and each entry is chained to the one before. Read more: Tamper-evident log
Token passthrough
Forwarding a token received for one service on to another. The MCP authorization spec forbids it, because it lets one leaked token open many doors. Read more: Token passthrough
Tool misuse
An agent using a legitimate tool in a harmful way, by mistake or because it was manipulated (OWASP ASI02). Read more: Tool misuse
Tool poisoning
Hiding instructions in a tool's description or results so an AI agent that reads them does something harmful. It's prompt injection delivered through the tool list. Read more: Tool poisoning
Unexpected code execution
An agent writing or running code that an attacker steered, leading to control of the machine it runs on (OWASP ASI05). Read more: Unexpected code execution
Untrusted content
Anything an AI agent reads that someone outside the business could have written, such as email, web pages, shared files and tool results. An agent treats it as data and never as instructions. Read more: Untrusted content

Keep reading

  • A heavy threshold gate lets one cleared path through while dozens of pink trails stop at it.

    Start here

    Agentic AI security

    Agentic AI security explained for owners: identity, least privilege, approvals and an audit trail for AI agents, from the team that runs them in Massachusetts.

    Brian Gagne

  • Two interlocking rings, a lock dial and an open aperture, joined by an amber clasp.

    Operations

    Agentic IAM and AI SOC in 2026

    Agentic IAM and AI SOC are the two 2026 security predictions that reach a small business. What they mean, what to ask your IT provider, and what to do first.

    Amelia Gagne

  • Medallions tangled into one pink knot, and one free medallion with its own amber thread.

    Identity and access

    Agentic IAM

    Agentic IAM gives every AI agent its own identity, scoped short-lived credentials and authority it can trace to a person. How it works and where to start.

    Brian Gagne

  • A pink stream of light is held at a round amber valve gate, waiting for a person's yes before anything leaves.

    Oversight

    Human approval for AI agents

    Human approval for AI agents that people actually read: which actions need a yes, how to avoid approval fatigue, and how standing approvals work.

    Amelia Gagne

  • A blurred hub drowns in pink cables while a clean hub in front holds one amber line.

    Identity and access

    AI agent permissions

    AI agent permissions decide what an agent can read, change and send. How to scope its tools, catch permission drift and run an access review.

    Brian Gagne

  • Hidden pink-lit devices behind dark panels, and one wide amber archway beside them.

    Threats

    Shadow AI

    Shadow AI is the AI your team uses without approval. How to find it without spying, write an AI policy people follow, and offer a better tool.

    Amelia Gagne

  • A sprawling swarm of small pink-lit tags, with one sealed and glowing amber inside a single capsule.

    Identity and access

    Non-human identities

    What a non-human identity is, why AI agents multiply them, and how to vault, scope and rotate API keys, tokens and service accounts before one leaks.

    Brian Gagne

  • A capsule travels through a line of rings and exits cleanly through an amber one.

    Identity and access

    AI agent lifecycle

    The AI agent lifecycle, run like staffing: who requests and owns an agent, what it gets on day one, how access is reviewed, and how to retire it cleanly.

    Amelia Gagne

  • Many pink cables converge on one hub, and every connection passes through its amber iris.

    Threats

    MCP security

    MCP security for businesses running AI agents: what the Model Context Protocol's authorization rules require, and how to vet servers and stop tool misuse.

    Brian Gagne

  • A column of stacked plates with pink-lit seams sealed by an amber top.

    Oversight

    Audit trails for AI agents

    An AI agent audit log proves what an agent did, who asked and who approved it. What to record, how to keep it tamper evident, and how long to keep it.

    Amelia Gagne

  • A smoked translucent slab carries a hidden pink thread toward the agent's tools, and an amber ring stops it before it reaches them.

    Threats

    Prompt injection

    Prompt injection turns an email, a web page or a file into orders for your AI agents. Why it can't be trained away, and the controls that limit the damage.

    Brian Gagne

  • Pink beads of personal detail drift between dark vessels, and one sealed amber vial holds only a few.

    Operations

    AI agents and personal data

    AI agents and personal data: what an agent should see, what it may keep, how to delete it on request, and how to log its work without hoarding personal details.

    Amelia Gagne

  • A swarm of pink alert shards streams through a dark sorting lens, while one amber lever stays locked until a person decides.

    Operations

    The AI SOC

    An AI SOC lets agents triage, enrich and summarize alerts. Containment, lockouts, deleting data and closing incidents still need a person to approve them.

    Brian Gagne

  • A column of vault tumblers lights pink at each seam, and the last aligns amber as the vault door edge begins to glow.

    Oversight

    AI agent security checklist

    An AI agent security checklist for owners: what to check before an agent acts, then identity, approvals, data, logging, incidents and retirement.

    Amelia Gagne

  • Ten black pillars veined with pink fractures stand in a row, and three amber rings thread through every one of them.

    Threats

    OWASP Agentic Top 10

    The OWASP Agentic Top 10, ASI01 to ASI10, explained for owners: each risk in everyday terms, a small-business example, and the control that answers it.

    Brian Gagne

  • Reference

    Questions

    AI agent security questions owners ask most: is it safe to let AI send email, what needs approval, who's responsible, and how Massachusetts law applies.

Most MSPs are resellers. Kief Studio is the manufacturer.

A reseller sends your problem to the company that made the software. We are that company: we build LTFI and run it, so the people you talk to can change it. Book a walkthrough and see it on your own business.

Nothing plays until you press play.