What this measures
How your security program governs, protects, watches, and recovers AI systems and agents. Two questions per area, six areas.
Six areas · 12 questions · organized by the six NIST CSF 2.0 functions
Could you prove your AI agents are under control?
Most teams know roughly where they stand on AI security. Almost none can demonstrate it. Twelve questions, and you leave with the list of what you could prove and what you could not, the kind of evidence an auditor or your board can ask for.
- 12
- questions
- 4
- levels
- 6
- areas
A self-assessment reflects what you report. The useful part is seeing which answers you could actually show.
You get your results as a one-pager too. Where should we send it?
In accordance with our Privacy Policy, you may be contacted about Trent AI or related products and services. You can unsubscribe at any time.
How it's scored
Each answer is 1 (Partial) to 4 (Adaptive). Levels adapted from the NIST CSF Tiers. Answers of 3 or 4 you couldn't show this week count one level lower on your effective score.
What you get
Your biggest gap in plain words, which answers you could and couldn't show, and a one-pager with your six area scores and what proof would look like for each.
The six areas
Six questions to ask about your AI agents
Govern
Who can shut an agent off, and who is accountable when it acts?
Ownership and policy. Risk tiers and approval.
Identify
Do you know what AI is running, and what it can actually reach?
Inventory of systems and agents. Threat modeling.
Protect
Your agents can probably reach more than they should.
Agent identity and access. Tool and action controls.
Detect
If one of your agents did something surprising tomorrow, would you even see it?
Misuse and anomaly detection. Activity logging.
Respond
An agent is doing something it shouldn't, right now. What happens next?
Incident plan. Kill switch.
Recover
After an AI incident, could you get back to a known-good state, and would anything change?
Restore and rollback. Lessons learned.
Sample result
What you see at the end
Biggest gap: Detect
If one of your agents did something surprising tomorrow, would you even see it?
You said you have this in place for 4 areas, but couldn't show it for 2.
| Domain | What you said | Could you show it? |
|---|---|---|
| Govern | Repeatable | Shown |
| Identify | Repeatable | Shown |
| Protect | Repeatable | Not shown this week |
| Detect | Adaptive | Not shown this week |
| Respond | Risk-Informed | Shown |
| Recover | Partial | Shown |
EffectiveClaimed
Why this exists
Yesterday's controls were not built for systems that act on their own.
Traditional security assumed predictable systems and human-driven actions. AI systems act autonomously, run on non-human identities, take real-world actions on untrusted input, and operate without human approval at each step. Policies that say "a human reviews every action" do not survive contact with a busy team. This check asks what is actually in place.
Standards anchor
Organized by the six NIST CSF 2.0 functions and aligned to the OWASP Top 10 for Agentic Applications. Also referenced: NIST AI RMF, CSA AICM, and the EU AI Act.
Who this is for
Security leaders at larger companies, and the people stuck owning AI security there, whether that's a team of five or the staff engineer who got handed it.
FAQs
How is the ASMM different from other AI security frameworks?
The ASMM is designed to assess the organization's security program, for its ability to govern, protect, detect, respond to, and recover from risks across AI adoption, development, integration, and operation, including agentic AI systems that reason, act, and operate with autonomy. It also assesses the security program (policies, governance, processes), not the technical posture of individual systems.
How is it different from NIST AI RMF or ISO 42001?
NIST AI RMF provides governance principles. ISO 42001 is a certifiable management system standard. The ASMM is an assessment. It gives you a score per area so you can track improvement over time, and shows where your program aligns with those frameworks and where it does not.
Does it apply to all AI systems or only agentic ones?
It applies to AI systems in scope, then uses the Autonomy Test to determine which systems qualify as agentic for agentic-specific control depth. The framework is designed for assessing and improving the security maturity of organizations developing or operating agentic AI systems, that reason, plan, and act autonomously, with reduced or risk-based depth for lower-risk or human-assisted AI systems, as appropriate.
How do I assess my organization's AI security posture?
Answer the 12 questions here, two per area. Each is scored 1 (Partial) to 4 (Adaptive), and answers you couldn't show this week count one level lower. Your area scores and biggest gap tell you where to start. The full model, 28 categories across the same six areas, is run with our team on a call.
Can I use this alongside NIST CSF 2.0?
Yes. The ASMM is aligned with the NIST CSF 2.0 structure. Same six functions (Govern, Identify, Protect, Detect, Respond, Recover). It complements NIST CSF 2.0 by applying its six-function structure to AI security maturity, including agentic AI-specific risks, rather than replacing it.
We already use a vendor AI security tool. Does this replace it?
No. ASMM is an assessment framework, not a security tool. It tells you where your security program stands and what to prioritize. Your existing tools are how you execute on that. Most organizations find ASMM useful precisely because it gives them a vendor-neutral baseline before or during vendor evaluation.
Doesn't NIST SSDF, OWASP SAMM, or the NIST AI RMF already cover this?
Partially, but each stops short of where agentic AI risk actually lives. SSDF and SAMM secure the humans building software. The NIST AI RMF governs AI as a system component. None of them address AI agents as autonomous actors or in systems that make runtime decisions, invoke tools, access data, and spawn sub-agents without human approval on every action. ASMM is built for that inversion. It inherits from CSF 2.0 and AI RMF for structural familiarity, then adds the six domains, four maturity tiers, and control evidence needed to assess and improve an organization's readiness to secure autonomous agent development specifically.
Does this assessment cover AI governance?
Yes. This assessment covers all six areas, including Govern, with two questions on AI security ownership and policy, and risk tiers and approval.
Why do you ask what I could show?
Because every maturity score is self-reported. A policy in someone's head and an approved policy with a version date score the same on most assessments. Here they do not. Answers of 3 or 4 you couldn't show this week count one level lower.
How is this different from the full ASMM?
This quick check takes two questions per area so you can see your biggest gap. The full model, 28 categories across the same six areas, is run with our team on a call.
What happens on the call?
We go through the full framework with you, starting with the answers you couldn't show yet, and you leave with a written list of next steps.
See which answers you could prove.
Twelve questions. Your biggest gap in plain words. A one-pager you can forward.