AI Assurance Intelligence Navigator

AI Assurance Intelligence Navigator: What It Is and How to Use It.

The Security of AI™ AI Assurance Intelligence Navigator is an interactive decision-support tool designed to help users determine what evidence can be used to evaluate whether AI security defenses are working as intended.

Identifying risks, threats, vulnerabilities, and mitigations is essential, but those activities leave an important question unanswered:

What evidence demonstrates that the defenses work?

The AI Assurance Intelligence Navigator helps answer that question by connecting AI security concerns with practical assurance activities, testing methods, evidence, artifacts, and ongoing monitoring.

The Navigator contains 49 assurance activities organized across AI assurance domains including:

• Model
• Data
• Application
• Agent
• Infrastructure
• Governance
• Operational

Each assurance activity begins with a practical assurance question and identifies the evidence that can be produced to support an assurance claim.

The objective is not to declare an AI system “secure” or eliminate uncertainty.

The Navigator provides a Security of AI™ assurance and evidence layerthat helps answer a different question:

What evidence do we have that the controls and defenses are actually working?

Why It Matters

AI security cannot end when a control is implemented.

A model may have adversarial-input protections. An AI agent may have restricted permissions. Sensitive data may have access controls. An application may include prompt-injection defenses.

But implementing a defense does not automatically demonstrate that the defense is effective.

Assurance requires evidence.

That evidence may come from testing, technical measurements, configuration verification, runtime monitoring, audit records, red-team activities, independent evaluation, or other forms of verification.

The AI Assurance Intelligence Navigator helps users examine:

What are we trying to demonstrate?

What testing or evaluation should be performed?

What evidence should be produced?

What artifact should document the results?

How often should the activity be performed?

What vulnerabilities and attack techniques does the activity address?

What frameworks relate to the assurance activity?

How much confidence does the evidence provide?

The goal is to move AI security beyond simply implementing controls toward evidence-based assurance.

How to Use the AI Assurance Intelligence Navigator

Step 1 — Explore the Assurance Activities

Select an assurance activity from the Navigator.

Activities are organized across AI assurance domains including:

• Model
• Data
• Application
• Agent
• Infrastructure
• Governance
• Operational

You can also use the search and filtering controls to narrow the displayed assurance activities.

Step 2 — Select an Assurance Activity

Select any activity to open its detailed assurance panel.

Each activity is centered on a specific Assurance Question.

For example, an assurance activity may ask whether there is sufficient evidence that a model can resist specified adversarial inputs, that model integrity has been maintained, that agent permissions are properly constrained, or that security controls continue to operate as intended.

The assurance question establishes what the activity is attempting to evaluate.

Step 3 — Review the Evidence Produced

Each assurance activity identifies the type of evidence that should result from the activity.

Evidence may include:

• Test results
• Security evaluation results
• Configuration records
• Runtime telemetry
• Audit records
• Red-team findings
• Technical measurements
• Verification results
• Human review
• Technical artifacts

The purpose is to make assurance observable and documentablerather than relying only on statements that a control has been implemented.

Step 4 — Review the Testing or Evaluation Method

The Navigator describes a practical method for performing each assurance activity.

Depending on the activity, this may involve techniques such as:

• Adversarial testing
• AI red teaming
• Model integrity verification
• Prompt-injection testing
• Data integrity testing
• Agent permission testing
• Tool-use testing
• Access-control verification
• Runtime monitoring
• Configuration assessment
• Security testing
• Audit and governance review

Where appropriate, the Navigator also identifies tools and techniques that may assist with the activity.

Tool references are provided for educational and decision-support purposes. Users should verify current versions, licensing, capabilities, and suitability for their environment.

Step 5 — Identify the Assurance Artifact

Testing should produce something that can be reviewed.

Each activity therefore identifies an expected Artifact Produced.

Examples may include:

• Test reports
• Evaluation results
• Audit records
• Monitoring reports
• Configuration evidence
• Security assessment results
• Red-team reports
• Verification records

These artifacts can contribute to the evidence needed to support security engineering, risk management, governance, acquisition, authorization, audit, or other organizational decisions.

Step 6 — Consider Testing Frequency

AI systems change.

Models are updated. Data changes. Applications are modified. Agents receive new tools. Permissions change. New vulnerabilities are discovered. Threat techniques evolve.

For this reason, assurance should not always be treated as a one-time activity.

The Navigator identifies assurance activities as appropriate for:

Continuous Monitoring

Activities that may require ongoing observation or automated verification.

Periodic Testing

Activities performed at established intervals to confirm that defenses continue to operate as intended.

Event-Driven Testing

Activities triggered by significant changes such as a model update, configuration change, new vulnerability, new threat intelligence, incident, or major deployment.

The frequency controls allow users to explore assurance activities according to these different approaches.

Step 7 — Understand the SOAI Assurance Measures

The Navigator provides three Security of AI™ decision-support measures.

Evidence Strength

Evidence Strength represents the relative rigor and repeatability of the evidence produced by the assurance activity.

Coverage

Coverage represents the breadth of the relevant attack surface or assurance concern addressed by the activity.

SOAI Assurance Confidence

SOAI Assurance Confidence combines Evidence Strength and Coverage to provide a relative indication of the confidence contributed by the assurance activity.

These measures are Security of AI™ analytical assessments.

They are not NIST, MITRE, OWASP, regulatory, certification, compliance, or vendor scores.

They are intended to support comparison, prioritization, and analysis within the context of the specific AI system.

Step 8 — Examine Framework and Vulnerability Relationships

Where applicable, the Navigator correlates assurance activities with recognized AI security and risk-management resources.

These may include:

AI Vulnerabilities

Relevant vulnerabilities, weaknesses, or risk conditions that the assurance activity may help evaluate.

MITRE ATLAS™

Relevant adversarial AI techniques associated with the activity.

NIST AI Risk Management Framework

Relevant AI risk-management functions, categories, or considerations.

OWASP

Relevant LLM, Generative AI, application, or agentic AI security risks.

These relationships help users move from a vulnerability or threat to the assurance activities that can produce evidence about whether associated defenses are effective.

Authoritative identifiers should be verified against their respective official sources.

Step 9 — Review Related Assurance Activities

AI assurance activities rarely stand alone.

For example, model integrity verification may need to be combined with access-control verification, supply-chain security, deployment verification, monitoring, and incident-response capabilities.

The Navigator identifies related assurance activities where appropriate.

Following these relationships helps users move from individual tests toward system-level assurance.