The twelve controls, in full

What each control actually requires.

Every control referenced by the assessment and the risk profiler, expanded: what it's for, why it matters, how to implement it, and what evidence an auditor would expect to see. See the methodology page for how these get selected per risk tier.

New: Five Rules for Agent Identity — the plain-English version of everything below.

Read the five rules →
INV

Discovery & inventory

INV-01Central agent registerMust from Contained
Objective

Every AI agent that exists is recorded in one authoritative list — not scattered across team spreadsheets, or existing only in someone's memory.

Why it matters

You cannot govern, secure, or account for an agent you don't know exists. This is the control everything else depends on.

Implementation

Register each agent at the point of deployment, not during a periodic audit. Capture name, purpose, environment, owner and technology at minimum.

Evidence an auditor would expect

A register export showing every entry with a named owner and a last-reviewed date.

Framework mappings
IMDADimension 1 — Assess and bound the risks
MASAI risk management — AI usage identification and inventory
NISTMAP
ISO/IEC 42001Clause 6.1 / Annex A.4
EU AI ActArticle 26 — Deployer obligations
INV-02Shadow AI detectionMust from HighRecommended from Contained and Elevated
Objective

Agents deployed without security review are discoverable — not just the agents someone remembered to register.

Why it matters

A register only covers what people choose to add. Detection catches what they didn't.

Implementation

Network or SaaS discovery tooling, API gateway logging, or a lightweight scan for common agent frameworks and API key patterns.

Evidence an auditor would expect

A record of at least one detection sweep and its findings, with follow-up actions logged.

Framework mappings
IMDADimension 1 — Assess and bound the risks
MASAI risk management — AI usage identification and inventory
NISTMAP
ISO/IEC 42001Clause 6.1 / Annex A.4
EU AI ActArticle 26 — Deployer obligations
IDN

Identity & attribution

IDN-01Unique agent identityMust from Contained
Objective

No agent shares credentials or an account with another agent, service, or person.

Why it matters

Shared identity is the fastest way to lose the ability to attribute an action to a specific agent — and the fastest way for one compromise to become hundreds. See the Salesloft-Drift breach below.

Implementation

Provision each agent its own service identity at creation. Reject requests to reuse an existing credential for a new agent.

Evidence an auditor would expect

An identity provider export showing a one-to-one mapping between agents and credentials.

Framework mappings
IMDADimension 2 — Make humans meaningfully accountable
MASScope and AI oversight
NISTGOVERN
ISO/IEC 42001Annex A.9 — Responsible use
EU AI ActArticle 14 — Human oversight
IDN-02Named human principalMust from Contained
Objective

Every agent has a specific, named human accountable for its purpose and continued need to exist — not a team, not a role. A person.

Why it matters

This is the invariant the product is built around. A distribution list is not accountable; a named person is — and it's enforced at the database layer, not just asked for on a form.

Implementation

Capture the principal at registration. Require it before approval. Re-confirm it at every recertification.

Evidence an auditor would expect

The Agent Passport's own principal field, plus the approval record showing it was checked before sign-off.

Framework mappings
IMDADimension 2 — Make humans meaningfully accountable
MASScope and AI oversight
NISTGOVERN
ISO/IEC 42001Annex A.9 — Responsible use
EU AI ActArticle 14 — Human oversight
ENT

Entitlement & least privilege

ENT-01Least-privilege entitlementsMust from Elevated
Objective

An agent holds only the permissions its stated task requires — not the broadest role available.

Why it matters

The agent's blast radius is its entitlement set. Overprovisioning turns a routine prompt injection into a serious compromise.

Implementation

Scope entitlements at deployment. Compare granted access to actual usage on a defined cycle, and revoke what's unused.

Evidence an auditor would expect

An access review showing granted-versus-used permissions per agent, with remediation tracked for gaps.

Framework mappings
IMDADimension 3 — Technical controls and processes
MASAI lifecycle management
NISTMANAGE
ISO/IEC 42001Annex A.6 — AI system lifecycle
EU AI ActArticle 15 — Accuracy, robustness, cybersecurity
ENT-02Pre-deployment blast-radius assessmentMust from HighRecommended from Elevated
Objective

Before an agent goes live, its tool bindings and data classifications are documented and its blast radius is rated.

Why it matters

This is the assessment that catches an entitlement problem before deployment, rather than discovering it after.

Implementation

A short, standard checklist completed at registration: which tools, which systems, which data classes, reversibility of actions.

Evidence an auditor would expect

The completed assessment attached to the Agent Passport.

Framework mappings
IMDADimension 3 — Technical controls and processes
MASAI lifecycle management
NISTMANAGE
ISO/IEC 42001Annex A.6 — AI system lifecycle
EU AI ActArticle 15 — Accuracy, robustness, cybersecurity
CRD

Credential handling

CRD-01Vaulted, short-lived credentialsMust from Elevated
Objective

No agent's credential lives in a config file, environment variable, or as a long-lived static token.

Why it matters

This is the exact failure mode behind the Salesloft-Drift breach — stolen OAuth tokens with no rotation reached over 700 downstream organisations. See below.

Implementation

Issue credentials from a vault, per-task where feasible, with automatic rotation.

Evidence an auditor would expect

A vault audit log showing credential issuance and rotation events for the agent.

Framework mappings
IMDADimension 3 — Technical controls and processes
MASAI lifecycle management
NISTMANAGE / COSAiS overlays (in development)
ISO/IEC 42001Annex A.6 — AI system lifecycle
EU AI ActArticle 15 — Accuracy, robustness, cybersecurity
CRD-02Independent revocationMust from High
Objective

A single agent's access can be revoked without disrupting any other agent or system.

Why it matters

If revoking one agent means an outage decision, revocation becomes something people avoid rather than something people do.

Implementation

Never share a credential across agents. Test revocation as part of onboarding, not only during an incident.

Evidence an auditor would expect

A logged test revocation with confirmation that only the target agent lost access.

Framework mappings
IMDADimension 3 — Technical controls and processes
MASAI lifecycle management
NISTMANAGE / COSAiS overlays (in development)
ISO/IEC 42001Annex A.6 — AI system lifecycle
EU AI ActArticle 15 — Accuracy, robustness, cybersecurity
AUD

Audit & containment

AUD-01Append-only action logMust from ElevatedRecommended from Contained
Objective

What an agent actually did — not just what it was asked — is recorded somewhere it cannot quietly edit.

Why it matters

Prompts and outputs describe intent. Actions are what create liability. Log the actions.

Implementation

Log tool calls, writes and escalations to a store the agent itself has no write or delete access to. Review on a scheduled cycle, not only reactively.

Evidence an auditor would expect

A sample of the action log for one agent, plus a record of the most recent scheduled review.

Framework mappings
IMDADimension 3 — Technical controls and processes
MASAI lifecycle management
NISTMEASURE
ISO/IEC 42001Clause 9 — Performance evaluation
EU AI ActArticle 12 — Record-keeping
AUD-02Tested kill switchMust from HighRecommended from Elevated
Objective

A misbehaving agent can be stopped mid-task — and someone has actually tried it.

Why it matters

An untested stop control is a belief, not a control. The Replit incident happened despite an explicit freeze instruction — the instruction lived in a prompt, not in an enforced control. See below.

Implementation

Define ownership and a target response time. Run the test at least annually, or after any material change to the agent.

Evidence an auditor would expect

A dated test record: who ran it, how long it took, what happened.

Framework mappings
IMDADimension 3 — Technical controls and processes
MASAI lifecycle management
NISTMEASURE
ISO/IEC 42001Clause 9 — Performance evaluation
EU AI ActArticle 12 — Record-keeping
LFC

Lifecycle & recertification

LFC-01Scheduled recertificationMust from Contained
Objective

Every agent's entitlements and continued need are reconfirmed on a defined cycle — not left as a one-time decision.

Why it matters

Access granted once and never reviewed becomes standing privilege — the same problem user access reviews exist to catch, applied to non-human identities.

Implementation

A recertification campaign with owner attestation, on a cycle set by risk tier (more frequent for higher tiers).

Evidence an auditor would expect

A completed attestation record with the owner's confirmation and date.

Framework mappings
IMDADimension 2 — Make humans meaningfully accountable
MASScope and AI oversight
NISTGOVERN
ISO/IEC 42001Clause 10 — Improvement
EU AI ActArticle 26 — Deployer obligations
LFC-02Decommissioning triggersMust from CriticalRecommended from High
Objective

An agent is retired when its owner leaves or its purpose ends — both triggers, not just one.

Why it matters

Orphaned agents with standing entitlements are the non-human equivalent of a dormant admin account nobody remembered to disable.

Implementation

Tie decommissioning to HR offboarding events and to a defined purpose-review cycle, with verification that access was actually removed.

Evidence an auditor would expect

A decommissioning record showing the trigger, the action taken, and confirmation of removal.

Framework mappings
IMDADimension 2 — Make humans meaningfully accountable
MASScope and AI oversight
NISTGOVERN
ISO/IEC 42001Clause 10 — Improvement
EU AI ActArticle 26 — Deployer obligations

Risk model version 2026.1. Last reviewed July 2026. Framework mappings are indicative — confirm current text for your jurisdiction and sector.

© 2026 Aseem Mohan · Assessment · Methodology · Five Rules for Agent Identity · Privacy notice