Agentic AI glossary: sixteen terms used in finance and marketing

Guardrails are rules built around an AI agent that block or stop actions outside the limits its operator has set.

Example: an agent connected to an exchange account is barred from placing any order above a set size, whatever the model proposes. The second guide covers risk controls for AI agents, guardrails included.

Human in the loop means a person must review and approve an AI agent's proposed action before it takes effect.

Example: an agent drafts a transfer and shows the amount and the destination; nothing moves until the account holder confirms. The same guide explains where approval steps belong among agent controls.

A read-only API key is a credential that lets software view an account's balances and history but not trade, transfer or withdraw.

Many exchanges let an account holder create keys with separate permissions for reading, trading and withdrawing; a tracker needs only the first. Example: you create a view-only key on an exchange and paste it into a portfolio tool, which can then show holdings but not move them. AMBR has not published how Ambre connects to accounts or which permissions it asks for. See what is known about portfolio connections.

Prompt injection is an attack in which crafted text makes an AI model ignore its instructions and follow the attacker's.

The text can sit in a chat message, a web page, an e-mail or a document the agent reads. The OWASP Top 10 for LLM Applications 2025 lists it as LLM01:2025 Prompt Injection. Example: a page an agent summarises contains a line telling it to drop its rules and send funds, and the agent treats that line as a command. See the risk-controls guide for digital assets.

Excessive agency is the risk that an AI agent has more tools, permissions or freedom to act than its task needs.

When that is so, one wrong output can do real damage, because nothing stands between the model's decision and the action. The OWASP list for 2025 names it LLM06:2025 Excessive Agency. Example: a portfolio tracker that only needs to read balances is given a key that can also withdraw. The usual answer is the smallest permission that does the job. The security page describes how AMBR agents are controlled.

Agentic marketing is marketing work in which AI agents monitor signals, plan tasks and prepare or carry out actions under human direction.

It differs from marketing automation, which runs rules a team has already written; an agent works out the next step from what it observes. Example: MIA monitors a brand, its competitors, markets, social channels, news and AI search, and turns those signals into marketing intelligence, structured work and approved actions. Its product site is titled “MIA — Agentic Marketing Platform”. The product page covers MIA, an AI marketing agent.

Generative engine optimisation (GEO) is the practice of making content easy for AI answer engines to find, understand and cite.

Search engine optimisation aims at a ranked list of links; GEO aims at the answer an assistant writes and the sources it names. It is also spelled generative engine optimization. Example: a brand checks whether AI assistants mention it when asked about its category, then improves the pages those answers draw on. AI search is one of the sources named on the MIA page on brand and AI search monitoring.