AgentGoverning is available for media comment on AI agent governance, regulatory compliance, and platform accountability.
AgentGoverning is the independent, non-commercial standards body publishing governance standards for autonomous AI agents. The AGS v2.1 standard defines 792 governance dimensions across 81 landscapes, covering everything from operational boundary enforcement and adversarial resilience to human override capability and cross-border regulatory compliance. The standard is published under a Creative Commons licence, freely available, and cannot be influenced by commercial relationships.
The AGS v2.1 compliance leaderboard reveals that the current industry leader — Microsoft Copilot Studio — scores just 26% against the standard. The average across the five platforms assessed is 19%. One platform has achieved verified status: Agent Shield™, scoring 99.9% across all 792 AGS v2.1 dimensions. The governance gap is not theoretical: it is measurable, documented, and published. These are not aspirational targets — they are the minimum structural requirements for agents operating in regulated environments.
This matters now because AI agents are no longer experimental. They are executing transactions, making decisions with financial consequences, and operating across jurisdictions with binding regulatory obligations. The EU AI Act is entering enforcement. The FCA is scrutinising AI-driven financial services. SOX compliance demands auditable decision chains. Yet no major platform has implemented the governance infrastructure these regulations require. The gap between what agents can do and what governance exists to constrain them is growing — and it is growing faster than regulation can respond.
Every platform is assessed against all 792 dimensions using a 0–3 scoring scale. A score of 0 means the platform shows no evidence of addressing the governance requirement. A score of 1 means there is partial or informal evidence — the capability may exist in documentation or marketing materials but lacks structural enforcement. A score of 2 means the platform has implemented a measurable governance mechanism that addresses the dimension, but it has not been independently verified. A score of 3 means the governance mechanism is structurally enforced, independently verified, and meets the full requirements of the dimension.
The overall compliance percentage is the sum of all dimension scores divided by the maximum possible score (2,376), expressed as a percentage. This gives a single, comparable figure across all assessed platforms. Scores labelled “Estimated” are based on publicly available documentation, API specifications, and published technical architecture. Scores labelled “Verified” have undergone independent assessment including adversarial testing.
The complete AGS v2.1 specification — 792 dimensions, 81 landscapes, 180,767 words. Published under CC BY 4.0.
Current estimated and verified compliance scores for all assessed platforms, updated with each assessment cycle.
For media enquiries, regulatory comment, or expert analysis.
framework@agentgoverning.com