GovernorAIby SentinelLayer Request a briefing
Find and qualify
Discovery & Shadow AIInventory the AI activity nobody registered AssuranceEvaluate an agent snapshot before production
Decide and enforce
Policy & GovernanceKeep the policy engine you already run Runtime EnforcementDecide at the action boundary Inference & Model GovernanceGovern the model call itself Multi-cloudOne policy, each cloud's own controls
Prove and observe
Evidence & ComplianceTamper-evident record, mapped to controls Decision ObservabilitySee the decision behind the action
Risk and security
RiskScore the action, and act on the score ContainmentStop a runaway run before it lands Security architectureHow the control plane itself is built
By persona
SecurityPer-call authorization and stop controls GRCEvidence chains and control mapping Platform & AI EngineeringGovern without touching the agent
By risk
Shadow AIUnmanaged agents and connections Data leakageIn prompts, completions and tool calls Destructive actionsChanges to systems of record
By platform
AWS BedrockInbound path, guardrail attachment ServiceNowNow Assist discovery and scoped control SalesforceAgentforce and Einstein Bots, in-org MCPRegistry, proxy, argument-aware inspection
By cloud and runtime
AzureConnect, discover, govern at the gateway Google CloudValidated connect mode, bootstrap session DatabricksCluster-side egress, pinned at create Kubernetes & sidecarIn your cluster, on the agent host

Resources

What is live, what is gated, what is next.

Credibility is part of the product, so coverage is stated honestly. Nothing is published here until it survives the same diligence we ask of an agent — no headline rests on a claim without an implementation, an integration and live-environment evidence behind it.

READING TRACKS

Three tracks, depending on what you need to decide.

Start where your question is. Each track stands alone.

Track 1

Vision

Why action governance is a different problem from model safety, and why independence from the delivery stack is the thing that makes evidence credible.

For leaders framing the problem
Track 2

Architecture

Enforcement points, outcomes, policy engines and the evidence chain. How a decision is made, where it is applied, and what it leaves behind.

For architects and platform engineers
Track 3

Enterprise

Rollout, separation of duties, regulatory control mapping and what an assessor will actually ask to see.

For security, GRC and programme owners

THE CLAIM GATE

Every capability sits in one of four states.

This table is the reason the rest of the site can be read at face value.

StageWhat it coversState
LiveRuntime policy enforcement at supported enforcement points, fail-closed capability matrix, stop controls, approvals, data controls, shadow-AI discovery, hash-chained audit, tenant isolation, and the Permissions assurance path.Shipped
Per deploymentSigned Ed25519/Merkle evidence bundles, privacy-isolation evidence, remote OPA, and several provider-native integrations require explicit configuration and verification. Not default-on.Configure
EmergingAll six assurance domains have an implemented evaluator. Whether a domain can be measured is a property of how a given deployment is wired, computed per deployment and reported with the missing prerequisite named. An unmeasurable domain reports not_assessed.In progress
RoadmapBroader measured assurance coverage, external offline evidence verification, and deeper automated control derivation.Next
Why this table exists

A capability page that only lists strengths cannot be checked. This one names the state of each area, including the ones that are configuration-dependent or incomplete — so a reader can tell the difference between what ships, what must be switched on, and what is not there yet.

PUBLISHED

What has cleared the gate.

All twenty-seven posts are here — the sixteen written for this library, ten brought over from the standalone blog, and one written since. The eight that were held for requalification have been brought back through the same gate the rest passed: product naming corrected to what the platform actually ships, invented ratios removed, and latency framed as a design budget with the benchmark page as the authority on what is measured.

Track 1 · Vision

01Control the Limbs, Not the BrainWhy enforcement at the action layer is the only meaningful control. 02The Agentic Gap: Why Traditional Security FailsAPI gateways are intent-blind. LLM guardrails are execution-blind. 03From Chatbots to Agents: The Liability ShiftAutonomy moves legal and financial liability to the deployer. 04AI Safety Without Enforcement Is TheaterPrinciples and prompt filtering are not security controls. 05The Ethics of Agency: Defining Rules of EngagementGovernance granular enough to allow utility while preventing harm. 17Why AI Guardrails Are Not GovernanceShaping what a model says is not deciding what it may do. 22Automated Offense Has an Existence Proof. Defense Does Not.What the OpenAI–Hugging Face incident shows about which half of the loop is still manual. 23Where Guardrails Stop and Tool Calls BeginGuardrails screen language; a tool call is a different control surface. 24From Governing Reasoning to Governing ExecutionPhase one governs what a model thinks. Phase two governs what it does.

Track 2 · Architecture

06Zero Trust for AI Agents: The New PerimeterIn the agentic age the perimeter is the action, not the network. 11Why Deterministic Policies Matter in AI SystemsYou cannot secure a probabilistic system with another one. 07Intercepting Tool Calls: How the Middleware WorksWhere the middleware sits, and what it can act on. 08Designing for <10ms Enforcement in the Execution PathA design budget for the execution path, not a measured claim. 09Why Enforcement Must Run Inside the Customer EnvironmentWhy the decision has to happen in your perimeter. 10What 'Fail Closed' Actually Means for Autonomous AIWhat the system does when the policy path is unavailable. 18Kill Switch Design Patterns for Autonomous AIFour target scopes, an account-wide flag, and why a TTL matters. 19Rogue Agent Detection: A Technical Deep-DiveBehavioural drift against a per-agent baseline, read from the action stream. 20Governance-as-Code for AI AgentsPolicy in version control, diffed and tested before it ships. 25Zero Trust at the Tool Call: The Enforcement PointThe policy enforcement point maps onto the call, not the network hop. 26No LLM in the Governance LoopA verdict made by a model inherits that model’s failure modes. 27The Operational Cost of Failing ClosedThe correct default is not a free one. What it costs, and how to pay it.

Track 3 · Enterprise

16From Principles to Policies: Operationalizing GovernanceGovernance is an ongoing lifecycle, not a one-time project. 12Audit Is Not Enough: Preventative Controls for the SOCA log tells you what happened; a control decides whether it does. 13SOC2 and the AI Agent: A Guide to Passing AuditsMapping agent actions onto an existing control environment. 14ROI of Guardrails: Accelerating the AI RoadmapGovernance as the thing that lets autonomy ship, not the brake. 15Human-in-the-Loop Is a Feature, Not a CompromiseHolding the action open is a capability, not a failure to automate. 21The Hidden Cost of DIY AI GovernanceThe build estimate prices the interceptor, not the decade after it.

Requalification record: four posts carried invented figures (a 99/1 split, a 70% reduction) presented as measured outcomes; those now describe the design intent instead. Eight named the product by the company name rather than GovernorAI. Two carried a propagation claim the source does not substantiate. Nothing was published that the platform does not state. The five blog posts were held to the same bar: one described a four-level kill-switch hierarchy ending in a global scope, where pkg/types/event.go declares four target scopes — session, agent, tool and namespace — with account-wide as an orthogonal flag; and two stated latency as achieved rather than budgeted.