🧭 AMC Methodology

How Agent Maturity Compass evaluates AI agent trustworthiness.

Core Principle: Evidence Over Claims

AMC scores agents based on observed behavior, not self-reported claims. An agent that says "I'm secure" scores differently from an agent that demonstrates security through tamper-evident evidence.

The gap between what agents claim and what evidence supports is often 84+ points — the "documentation inflation gap."

Five Dimensions

Dimension	What It Measures
D1 — Governance	Charter, permissions, approval workflows, action policies
D2 — Capability	Task execution, autonomy duration, tool usage, delegation
D3 — Culture & Alignment	Safety, honesty, transparency, value coherence
D4 — Resilience	Error handling, adversarial robustness, drift detection
D5 — Skills	Memory, reasoning, calibration, factuality

Maturity Scale (L0–L5)

Level	Name	Description
L0	Absent	No evidence collected
L1	Initial	Some intent, little operational evidence
L2	Developing	Basic governance, partial coverage
L3	Defined	Systematic processes, evidence-backed
L4	Managed	Quantitative measurement, continuous monitoring
L5	Optimizing	Predictive, self-improving, externally validated

Evidence Trust Tiers

Tier	Weight	How
OBSERVED_HARDENED	1.1×	AMC-controlled adversarial scenarios
OBSERVED	1.0×	Captured via gateway proxy
ATTESTED	0.8×	Cryptographic attestation
SELF_REPORTED	0.4×	Agent's own claims (capped)

Assessment Coverage

244 default diagnostic questions across 5 layers, plus a free opt-in 20-question lifecycle expansion for 264 questions total

142 assurance packs — red-team attack simulations including alignment faking, sandbagging, and scheming detection

Research-backed scoring modules — evidence-weighted algorithms

8 compliance frameworks — EU AI Act, GDPR, ISO 42001, SOC2, NIST AI RMF, HIPAA, PCI DSS, ISO 27001

41 industry domain packs — health, education, finance, mobility, governance, technology, environment

Getting Started

curl -fsSL https://agentmaturity.co/install.sh | sh
amc                    # full score, no setup
amc domain pack access # planned tier status; public checkout is not live
amc assurance run --all # Red-team testing
amc proof check --manifest rules.json --input claim.json # Domain Proof Lane
amc comply report --framework EU_AI_ACT