ARM-B · Benchmark by Evolve Edge
ARM-B: The AI Risk Management Benchmark
The only independently validated benchmark for AI governance assessment accuracy
What ARM-B measures
Gate Classification accuracy
Accuracy of adopt, hold, and reject decisions when a case is evaluated against governance gates.
Route Assignment accuracy
Accuracy of routing each case to the correct review and delivery path for its risk profile.
Verification Logic accuracy
Accuracy of the checks that confirm whether required controls and evidence are actually satisfied.
Remediation Prioritization accuracy
Accuracy of ordering remediation work so the highest-risk gaps are addressed first.
Methodology
ARM-B runs a fixed case set against a pinned model with deterministic settings, so any result can be reproduced from the published manifest. The validator scored outputs without access to the gold labels used during development.
- Cases
- 100
- Validation
- Gold-blind (validator never saw training data)
- Model
- gpt-5-nano-2025-08-07 (pinned)
- Temperature
- 0
- Seed
- 42
- Manifest
- sha256:1a9e2126209d8bd8dc48f62c0b66e9231c2bca3b0e01a5ff3ac8a5f1c790c4da
Leaderboard
Results are anonymized by vertical.
| Vertical | Score | Cases Passed | Date Validated |
|---|---|---|---|
| AI-Native SaaSEvolve Edge · Benchmark Author | 100/100 | 100 | Sept 26, 2026 |
| Healthcare TechIllustrative, pending validation | 97/100 | 97 | Pending |
| FintechIllustrative, pending validation | 95/100 | 95 | Pending |
Rows marked illustrative are example entries and do not represent completed validations.
Request ARM-B Validation for Your Platform
Run your AI governance assessments through ARM-B and get an independently validated accuracy score you can share with buyers and auditors.
Request ARM-B Validation for Your Platform