Three interconnected properties — a standard, an audit, and a product — forming the infrastructure layer for trusted AI agent commerce. Each property reinforces the others. Together, they create a governance moat that compounds with every deployment.
AgentGoverning is the open governance standard that defines what it means for an AI agent to be trustworthy. AGS v3.1 specifies 841 measurable governance dimensions across 101 landscapes (Agent Shield ships 783 of them) across safety, alignment, transparency, accountability, and operational integrity — providing the common language the industry has lacked.
The standard is platform-neutral by design. Any platform, any framework, any agent architecture can be assessed against it. This is deliberate. A governance standard that only works with one product is not a standard — it is a sales tool. AgentGoverning aligns with 12 major regulatory and industry frameworks including the EU AI Act, NIST AI RMF, ISO/IEC 42001, and OWASP guidelines.
By publishing AgentGoverning as an open standard, Imperium establishes the taxonomy — the shared vocabulary — for AI agent governance. When enterprises, regulators, and insurers discuss whether an agent is “governed,” they will reference these dimensions. The standard does not compete with regulation. It translates regulation into measurable, auditable, machine-readable criteria.
The strategic value is structural. Whoever defines the standard defines the market. Every competitor that builds governance tooling must either adopt AgentGoverning or explain why their alternative is better — and against hundreds of dimensions with 12 framework alignments, that explanation becomes increasingly difficult to make.
Agent Aegis operates two separate audits: the LLM Audit tests governance platforms via adversarial attacks across the AGS dimensions, while the Agent Audit tests AI agent platforms for natural governability.
The LLM Audit does not ask agents whether they are safe. It attacks them across 9 distinct attack vectors — prompt injection, goal drift, authority escalation, data exfiltration, resource abuse, deceptive alignment, context poisoning, multi-agent collusion, and supply-chain compromise — and measures what survives.
Each submission is tested across the AGS governance dimensions. The results are cryptographically signed, timestamped, and immutable. An Agent Aegis score is produced through adversarial testing and cryptographically signed, so a submitter cannot self-declare it. The certificate is the cryptographic proof.
Agent Aegis operates on a per-agent submission model: each agent is assessed against the standard through adversarial testing and issued a signed result. Pricing is available on request — contact us.
The data asset is equally significant. Every submission generates a detailed adversarial profile — how the agent failed, where it was vulnerable, which attack vectors succeeded. This corpus of adversarial intelligence is proprietary and compounds with every assessment, and would be difficult to replicate without running an equivalent audit at scale.
Agent Shield is the real-time governance enforcement platform — the product layer that operationalises the AgentGoverning standard. It is not a monitoring dashboard. It is a constitutional governance engine that intercepts, evaluates, and enforces governance policies on every agent action, every transaction, every decision — in real time.
The platform was architected against the AGS governance dimensions; its earlier Agent Aegis LLM Audit score is withdrawn pending a re-measured run. It serves 1,372 API operations (all versions, including operator routes), providing granular governance control over agent identity, capability boundaries, financial limits, data access, inter-agent communication, and human escalation protocols. Seven UK patent applications have been filed covering the core architectural innovations; none has been granted yet.
Agent Shield is the commercial engine of the ecosystem. It is the subscription product built for enterprises to govern their AI agent fleets. But it is not a standalone product — it is the product expression of the entire Imperium stack. It implements AgentGoverning. It is audited by Agent Aegis. It enforces the standard that the ecosystem defines.
This integration is the moat. A competitor can build a governance product. They cannot simultaneously own the standard that defines governance, the audit that validates it, and the product that enforces it. Agent Shield does not compete on features. It competes on architectural position — it is the reference implementation of the governance standard, built to be audited by the adversarial audit.
When OWASP published its Top 10 web application security risks, it did not build a product. It defined the vocabulary. Every web security vendor, every penetration testing firm, every compliance framework that followed had to reference OWASP — because OWASP defined what “secure” meant. AgentGoverning occupies the same structural position for AI agent governance. It defines what “governed” means. The AGS dimensions become the lingua franca. Competitors must either adopt the standard or spend years building an alternative — years they do not have, because EU AI Act obligations are phasing in now, with high-risk duties from 2 December 2027, and enterprises need governance taxonomy now.
Under the EU AI Act, as amended by Regulation (EU) 2026/1744, transparency duties apply from 2 August 2026 and high-risk duties from 2 December 2027 (Annex III) and 2 August 2028 (Annex I). The UK has stood up its AI Security Institute (formerly the AI Safety Institute), and US AI policy continues to evolve at the federal and state level. The direction of travel is clear: enterprises deploying AI agents will increasingly need to demonstrate governance, accountability, and auditability. The question is which governance framework becomes the reference standard. AgentGoverning, with its hundreds of dimensions and broad framework alignment, is built to be comprehensive and available today.
Every Agent Aegis audit submission generates a detailed adversarial profile — how the agent responded to each of the 3 frontier LLMs, where it failed, where it succeeded, which governance dimensions were weakest. This data compounds with each assessment and informs how the governance standard and platform evolve. Submitter IP is protected, and results are not shared between submitters.
| Type | Asset | Description | Status |
|---|---|---|---|
| Patents | Agent Shield patent family | Seven UK patent applications covering AI agent governance methods and systems | UK IPO — Pending |
| Trademark | Agent Shield | Registered UK Trade Mark No. UK00004362695, owned by Andrew Hoyle — class 36 (financial risk management), class 42 (software as a service, AI platforms, software development) and class 45 (regulatory compliance auditing and advisory) | Registered |
| Standard | AgentGoverning™ | Open governance standard (CC BY 4.0) — AGS v3.1, 841 dimensions across 101 landscapes, 12 framework alignments, platform-neutral specification for AI agent governance | Published |
| Audit | Agent Aegis™ | Adversarial governance audit — 3 frontier LLMs, cryptographically signed scoring, adversarial validation methodology | Operational |
Patent applications pending with the UK Intellectual Property Office. Agent Shield® is a UK registered trade mark owned by Andrew Hoyle.
Imperium Business Intelligence has built the complete governance infrastructure layer — an open standard, an adversarial audit, and a working product — three properties that reinforce each other, covered by seven patent applications, tested adversarially, and being rolled out across multiple products, starting with AgentFlowMe.
This is not a feature. It is not a module. It is the infrastructure layer for trusted AI agent commerce. And infrastructure layers, once established, are extraordinarily difficult to displace.
The deliberate choice is not whether to govern AI agents.
It is whether to prove it, or ask to be trusted.
Agents are not software to be secured. They require governance infrastructure.