841 dimensions · 101 landscapes · Open specification · One reference implementation
The AGS standard is published by AgentGoverning. Agent Shield is the reference implementation.
Open standards allow regulators to audit against a known framework. They allow customers to verify claims independently. They allow the industry to build towards a common governance baseline.
The reference implementation — Agent Shield — is the reference implementation of the AGS standard.
Infrastructure is built on open standards. Always has been. The AGS standard is published by AgentGoverning, our open AI governance standards initiative (by Imperium Business Intelligence).
Openly specified so regulators can audit it. Built into a reference implementation so organisations can test it.
Read the standard. Verify the certificate. That is the governance chain.
Governance infrastructure should rest on an open standard. AGS is published under CC BY 4.0, and Agent Shield implements it.
AGS v3.1 is the foundational governance specification for autonomous AI agents operating autonomously. It defines 841 governance dimensions across 101 landscapes (Agent Shield’s machine-readable corpus ships 783 of them, AG-001 to AG-783), spanning authority controls, constitutional constraints, pattern monitoring, and audit integrity.
The standard establishes that AI agents are not software to be secured — they are operational entities requiring the same governance infrastructure as any operational entity within defined authority limits. AGS v3.1 sets out a governance baseline for AI agent operations: what an organisation should have in place before deploying autonomous agents in financial or regulated environments.
“If an agent action cannot be governed, it cannot be authorised.”
Agent Aegis is the adversarial audit published by AgentGoverning, which, like Agent Shield, is operated by Imperium Business Intelligence. It tests agent behaviour across the AGS dimensions, grouped into categories, and produces a letter grade from A+ (genuinely aligned) to F (deployment blocked).
Unlike safety audits that test what agents can do, Agent Aegis tests what agents will do when presented with realistic governance scenarios. It measures alignment, behavioural integrity, and governance compliance under adversarial conditions.
Network governance is the trust infrastructure for AI agent networks. In Agent Shield the trust registry, inter-agent handshake and audit federation exist in code and are switched off by default while they are completed. When agents from different organisations transact with each other, it provides the governance framework — the equivalent of what established trust frameworks provide for traditional commerce.
The standard establishes trust registries, inter-agent governance protocols, and cross-organisational audit chains. It ensures that governance doesn’t stop at the boundary of a single organisation but extends across the entire network of AI agents participating in autonomous operations. Think of it as the trust network for AI agent interactions. The standard does not tell organisations how to build agent systems. It defines the governance standard that allows organisations to transact with each other. Network governance does the same for autonomous AI agents across organisational boundaries.
“Governance that stops at your firewall is not governance. It is theatre.”
Designed so every cross-boundary transaction passes through the governance network
The AGS v3.1 standard is published as an open specification under CC BY 4.0. Anyone can read it, implement against it, or build tooling for it. The standard belongs to the industry. Agent Shield is the reference implementation. It does not yet implement every dimension: Agent Shield ships 783 of the 841, and the AGS benchmark tests 26 of those against the engine.
Request a compliance assessment and see how your agent fleet measures against the AGS v3.1 standard.
Published by AgentGoverning, our open AI governance standards initiative (by Imperium Business Intelligence, which also develops Agent Shield). Agent Shield is the reference implementation.