AGS Assurance Framework
Three tiers of governance assurance
Modelled on established accountancy assurance standards — from agreed-upon procedures through reasonable assurance.
Modelled on established accountancy assurance standards — from agreed-upon procedures through reasonable assurance.
The AGS Assurance Framework defines three tiers of governance assurance, each modelled on an established accountancy assurance engagement type. Tiers are cumulative — each builds on the evidence requirements of the tier below.
Adversarial testing of specific dimensions at a point in time. What the current benchmark produces. An independent assessment of selected governance dimensions using adversarial techniques, with results published on the leaderboard.
Design review across all dimensions. No material exceptions noted. Requires documented controls and protocol file coverage across all 796 AGS v2.2 dimensions.
Full operating effectiveness testing across all dimensions over a defined period. Requires test suite evidence, design documentation, and adversarial verification. The highest tier of AGS governance assurance.
Each tier builds cumulatively on the requirements of the tier below.
| Tier | Evidence Required |
|---|---|
| AGS-AUP | Adversarial benchmark results · Specific dimensions tested at a point in time |
| AGS-LA | All of AGS-AUP + Documented controls for all 841 dimensions + Protocol file coverage across every dimension |
| AGS-RA | All of AGS-LA + Test suite evidence (unit tests mapped to dimensions) + Continuous monitoring + Period of operation (minimum 90 days) |
Agent Shield is the first platform to achieve AGS-RA (Reasonable Assurance) across all dimensions of the AGS v2.2 standard.
Each tier has a distinct submission process. All submissions begin through the verification page.
Submit your platform for LLM Audit or Agent Audit. Adversarial testing is conducted against specific dimensions. Results are published on the leaderboard.
Submit documented controls and protocol files covering all 841 dimensions. AgentGoverning reviews design coverage and issues negative assurance if no material exceptions are found.
Submit all AGS-LA evidence plus test suite evidence mapped to dimensions, continuous monitoring configuration, and evidence of a minimum 90-day operating period. Full operating effectiveness testing is conducted.
Begin the assurance assessment process for your platform.