Agentic AI | AI Glossary | Prompt Security

Agentic AI

What Is Agentic AI?

An agent can plan a multi-step task, decide which tools or data sources to use, and execute and adapt its actions along the way, often without a human approving each step. This is no longer theoretical: coding agents, browser agents, and task-automation agents are already in production, frequently connected to internal systems through protocols like MCP. That autonomy is the point, and it's also what expands the attack surface. Every tool an agent can call and every action it can take without a human in the loop is a new place for something to go wrong. As agentic systems take on more real work, the security conversation shifts from "is the output correct" to "what did the agent actually do, and did anyone approve it."

How It Works / Why It Matters

Agents typically operate in a loop: observe, plan, act, observe the result, repeat, rather than a single request-response exchange.
Autonomy and oversight trade off directly against each other. The more steps an agent completes without a human checking in, the harder it is to catch a problem before it has already happened.
Agents are often connected to real systems (email, databases, code repositories, internal APIs) through MCP or similar protocols, so a manipulated agent isn't just generating bad text, it can take a real action.
Guardrails for agentic systems tend to focus on scoping what an agent is allowed to touch, not just filtering what it's allowed to say.

FAQ

No. A chatbot responds to what you type. An agent plans and executes a sequence of actions toward a goal, often across multiple tools, with much less back-and-forth confirmation from a person.

No, but MCP has become a common way agents connect to external tools and data, which is why the two topics come up together so often.

Reduced oversight. Traditional AI risks like prompt injection or hallucination become more serious once an agent can act on a bad instruction or a wrong fact instead of a human just reading it and moving on.

Share this page

Related Terms

Model Context Protocol (MCP)

Model Context Protocol (MCP) is an open standard, originally developed by Anthropic, for connecting LLMs to external tools, systems, and data sources through a single interface rather than custom integrations for every connection.

Privilege Escalation

As LLMs and agents connect to more internal systems and MCP servers, the risk grows that those integrations are used, deliberately or not, to gain access beyond what was intended.

Shadow AI

Shadow AI describes the AI tools and agents employees adopt on their own, from chat assistants to coding copilots to autonomous agents, without visibility or approval from IT or security teams.

The Agentic AI Attack Surface: Where Risk Lives Beyond the Prompt

What OpenClaw's (Clawdbot) Virality Reveals About the Risks of Agentic AI

ClawSec

OneClaw

Prompt for Agentic AI: Guardrails for Autonomous Systems

Securing OpenClaw: AI Agent Security Risks

Is agentic AI the same as a chatbot?

Does agentic AI require MCP specifically?

What’s the biggest security risk specific to agentic AI?