The Elmoz Academy. AI agent and non-human identity security, explained.

Every page here answers one question about what AI agents and machine identities can reach, and what to change when the answer is too much. Written from incident teardowns, not from product copy.

Book a demoRead the attack paths guide

BROWSE

Five categories.

Start where your question sits. Every category is an index of pages, not a landing page.

Non-human identitiesService accounts, API keys, OAuth apps and the agents that run on them.Browse topics →AI agent securityWhat an agent can do on its own, and the access that makes it possible.Browse topics →MCP securityModel Context Protocol servers, the credentials behind their tools and what those reach.Browse topics →

Attack paths and reachHow single access steps connect into a path from an agent to sensitive data.Browse topics →Governance and regulationWhat the EU AI Act, NIS2 and the OWASP agentic list actually require of agent access.Browse topics →

ANATOMY

What every page gives you.

A definition you can quote.Two sentences at the top, written so they can be lifted straight into a ticket, a policy or an answer to an auditor.

The mechanism, step by step.

One table you can act on.

Four steps to put it into practice.

The questions people actually ask.

Page anatomy

TakeawaysFour statements, before anything elseTop

DefinitionTwo sentences, quotable

MechanismNumbered sections

TableOne, with the rows that matter

In practiceFour steps

FAQThree questions

Every page, in one list.

25 pages today. The list below is the full index.

OAuth tokens and AI agentsWhy a grant is standing access, and how to review grants nobody owns.Read the page →Secrets sprawl and long-lived tokensThe five places copies end up, and a rotation order that works.Read the page →Machine identity vs non-human identityThree overlapping terms, and the one distinction that changes the controls.Read the page →

What is a non-human identity (NHI)?The definition, the five types and why every control built for people fails on them.Read the page →NHI security: 7 risks and the fix for eachStatic keys, missing owners, grown permissions, shared identities and stale grants.Read the page →What is an OAuth grant?Standing delegated access to a user account, granted once and rarely reviewed again.Read the page →

Agent-to-agent delegation risksConfused deputy in agent chains, and whose access is used at the end.Read the page →How to discover AI agents in an enterpriseFour credential trails to follow, what each returns and what it misses.Read the page →AI agent audit trail and loggingThe five questions asked after an incident, and what answers them.Read the page →

AI agent inventory: what to trackThe four places agents leave traces and the eight fields worth tracking.Read the page →What is a shadow AI agent?An agent with real access that nobody registered, reviewed or owns.Read the page →AI agent permissions: least privilege for agentsWhy an agent uses every permission it holds, and four ways to scope one.Read the page →

What is least agency?The principle of limiting what an agent may decide, not only what it may access.Read the page →MCP server review and allowlistingFive questions before approval, and what to do with the ones already installed.Read the page →MCP server securityThe server is an identity, and every agent calling it inherits its reach.Read the page →

What is an MCP server?The parts it is made of, and where its credentials actually sit.Read the page →What is agent reach?The four layers of what an agent can get to, and how to measure them.Read the page →What is an attack path?A chain of access steps from an entry point to something that matters.Read the page →

AI agent blast radiusWhat one compromised agent would affect, before anyone stops it.Read the page →What is a breakpoint?The single change that closes a path with the least disruption.Read the page →What is distance to production?How many steps sit between an agent and a destructive action.Read the page →

ISO 27001 and non-human identitiesFive Annex A controls, the evidence auditors ask for, and where the standard is silent.Read the page →EU AI Act and AI agentsWhat the Act asks of agent access: classification, logs, oversight, deployer duties.Read the page →NIS2 and non-human identitiesWhere machine identities sit in Article 21, and when the 24-hour clock starts.Read the page →

OWASP for agentic AIThe agentic risk list read as an access problem, and where it stops being useful.Read the page →

FAQ

Questions, answered.

The two questions that decide which category you need.

What is AI agent security?

AI agent security is the practice of controlling what AI agents can access and do: which identities and credentials they use, what data they can reach and which actions they may take on their own. Most of it is access control rather than model behaviour.

What is the difference between NHI security and AI agent security?

Non-human identity security covers all machine identities such as service accounts, keys and OAuth apps. AI agent security focuses on agents, which are machine identities that also decide their own actions, so every permission they hold becomes a permission they may use.

Beyond the Academy.

Where the research and the product live.

ResearchIncident teardowns: what the agent reached and what to check.Read the research →Attack paths guideThe long form guide, start to finish.Read the guide →The platformDiscover, trace, prioritize and fix, in one security graph.See the platform →About ElmozWho writes this and why.Read about us →

See what your agents can reach. Before someone else does.

Book a demo and we walk through the access paths in an environment like yours.

Book a demoRead the attack path guide