Home / Resources / How AIBound grades AI
Guide
How AIBound grades AI from A to F
AIBound gives each AI it finds a grade from A to F. The grade shows how much harm that AI could do on an employee device, so the worst ones are handled first. This guide explains what gets a grade, what a grade is based on and what happens after it.
Version 1 · Published 17 November 2026 · Updated 17 November 2026
What is inside this guide
- What gets a grade
- What a grade is based on
- How findings connect to the OWASP lists
- What happens after a grade
- What a grade is not
Who this guide is for
- CISOs and security leads: see how a grade is reached before you rely on it.
- Security engineers: see why an AI got its grade and what evidence sits behind it.
- GRC and audit teams: see how grades turn into records you can show an auditor.
What gets a grade?
AIBound grades each AI it finds on an employee device. That covers AI agents, personal agents, MCP servers, skills, sub-agents, models and apps.
Each one gets its own grade, so a long list becomes a ranked list. The security team starts with the few that could do the most harm.
What is a grade based on?
A grade looks at three things: what the AI is, what it can reach and how it behaves. A skill that tells a coding agent to read credentials is judged differently from a note-taking app that reaches nothing.
For skills, AIBound reads each one the way an attacker would. It combines fixed checks with a layered AI review that checks its own verdicts. Every finding points to the exact line, so your team can see why a skill got its grade.
How do findings connect to the OWASP lists?
OWASP is a nonprofit that publishes lists of the main security risks for AI. Two of them matter here: the OWASP Top 10 for LLM Applications and the OWASP MCP Top 10. AIBound tags its findings with these lists, so the evidence uses a standard vocabulary and not a private one.
Three examples show how a finding lines up with the lists:
- A skill hiding instructions that make an agent do something its user did not ask for is a Tool Poisoning case (MCP03).
- An MCP server that nobody approved or listed is a Shadow MCP Servers case (MCP09).
- An agent with far more access than its job needs is an Excessive Agency case (LLM06).
The OWASP MCP Top 10 is still a beta release at the time of writing, so its numbering may change.
What happens after an AI gets a grade?
The grade decides which rule applies. You set the rules, and each action is allowed, put to the employee as a question, or denied. For example, a policy can deny any skill graded D or F at install.
The grade, the evidence behind it and the decision are kept together. Each risk report shows the grade, the evidence and a recommendation, and it can be downloaded for audit. AIAssure uses the same records for EU AI Act, NIST AI RMF and ISO/IEC 42001 evidence.
What is a grade not?
A grade is a guide. The rule that stops a high-risk AI is what protects the company, and the grade tells you where to point it.
A grade is also not an AI risk assessment. A grade is quick, automatic and covers every AI found. An assessment is a deeper, written review of one AI system, and a grade helps you decide where to spend that time.
Grading is a normal part of this market. Gartner's public description of AI usage control says these tools discover AI use, assess risk and let security teams enforce policy.
Other resources
AI Usage Control (AIUC) requirements: a buyer's checklistFAQ
Frequently asked questions
What does an F grade mean?
F is the lowest grade. It marks an AI that AIBound judges to be among the most risky, so it goes to the top of the review list. What happens next is your policy: allow, ask or deny.
Does a grade change over time?
Yes. AI changes as people update it, so the grade is not fixed. AIBound keeps reading what is installed, and the grade follows what it finds.
Where can I see the evidence behind a grade?
Each AI has a risk report with its grade, the evidence behind it and a recommendation. You can download the report for an audit.
Is an AI risk grade the same as an AI risk assessment?
No. A grade is a quick ranking of every AI found. An assessment is a deeper written review of one AI system or one use of AI. Grades tell you which AI deserves an assessment first.
Sources referenced
Where the facts on this page come from. The facts about AIBound come from AIBound's own product pages.
- Source 1. OWASP Foundation. OWASP MCP Top 10. Checked 17 November 2026.
- Source 2. OWASP Gen AI Security Project. OWASP Top 10 for LLM Applications. Checked 17 November 2026.
- Source 3. Gartner. Market Overview for AI Usage Control (abstract). Checked 17 November 2026.