User Guide

Getting started

Go to autogovern.io/app — the Governance Workbench. No sign-up is required for any tool there; it's free to use and runs immediately.

A short onboarding flow asks three quick questions — what kind of AI system, what area it operates in, what you want to do first — and jumps you straight to the most relevant tool.

You can always skip it and browse the available tools yourself.

Find tools by task

The Workbench menu groups independent tools by task. Start and manage systems is open initially; select any group to expand it. A direct tool link opens its group automatically. Choose Show all tools to expand every available group, or use Search tools and pages. Recent tools appear while you work in the current browser tab.

• Start and manage systems: System Intake, AI Registry, Workspace inventory and Discovery shortcuts.
• Assess and test: fairness, drift, documentation, LLM safety, vendor risk, red teaming, artifact scanning and threat analysis.
• Build governance: organization governance, model cards, crosswalks, vendor answer packs and Governance Copilot.
• Monitor and investigate: Incident Intelligence, Shadow AI Detection, Reg Timeline and Regulatory Horizon.
• Analyze and simulate: Governance Agent and Digital Twin & Simulation. These tools analyze or simulate; they do not apply changes to your systems.
• Prepare evidence: Compliance Dossier and Evidence Rooms.
• Specialist tools: transparency, model risk, fair lending, financial services controls and agent handshakes.

Private platform operations remain available only through authorized administration. They are not part of the public tool catalogue.

Browse all tools and their direct links →

Website navigation

Platform contains Workbench, Discovery & Integrations, Evidence Rooms, Start here and Private Governance Studio. Solutions contains industry workspaces and guides. Regulations & Frameworks contains regulations, deadlines and standards. Learn contains guides, templates, the blog and research. Services contains advisory services and Contact.

Search opens common tasks first. Type a task or tool name; exact title matches come first. Command+K or Ctrl+K opens search, arrow keys browse, Enter opens a result, and Escape closes it. Use Browse all tools and resources for the complete catalogue. On mobile, open the menu and expand a category; Platform starts expanded. Inside the Workbench, Choose a tool opens the tool list and privacy options. Choosing a tool closes that list so its content is immediately visible.

Workspace opens your organization. Its section links lead to AI systems & inventory, Saved work & insights, and Reports & effectiveness. Expand Administration for Organization & team or API access & audit.

Policy Compiler

Administrators: open Private admin → Regulatory review & platform operations → Policy Compiler in the Workbench. Write a governance rule the way you'd say it out loud — "Customer-support agents may read account data but may not change payment details without human approval" — and Compile policy converts it into a structured rule:

• Who it applies to
• What resource it covers
• Which action types
• A decision — allow, review, or deny

Review and edit every field before saving; nothing is guessed silently. Save it as a draft, then Simulate it against generated test cases (or your own) to see exactly what it would decide before it affects anything real.

Only active policies are consulted by the Agent Control Plane — activating checks for conflicts with other active policies covering the same ground with a different decision first. Edits are saved as new versions using the existing family ID. Rollback restores the approved previous version.

Review and activation lifecycle

Open Workbench → Policy Compiler. The compiler registry governs the shared demo workspace. Platform administrators manage its lifecycle; organization owners are not automatically platform administrators.

1. Compile, inspect, and Save as draft. To revise an existing family, a platform administrator supplies its family ID in the editor.
2. Use Inspect version & review history and Simulate. Choose Submit for review and record the reason.
3. A different platform administrator inspects the exact version and chooses Approve version or Request changes, with a reason. Submission alone does not activate anything.
4. After approval, choose Activate approved version. Conflicts require an explicit override and recorded reason. Changes to reviewed content invalidate approval.
5. Deprecate stops that version from applying. Rollback to approved predecessor requires the previous version to have its own valid approval and passes the same conflict check. Older versions remain visible for review.

Existing active policies keep running. Review history records actor IDs, reasons, timestamps and conflict overrides; it is available to platform administrators. Remediation Center policies retain their separate remediation approval workflow. Environment/jurisdiction targeting, staged rollout and OPA/Cedar export are not included in this release.

The full set of active policies is available as a signed, machine-readable bundle at GET /api/policy-bundle for any external system to fetch and verify — the same cryptographic key that signs Trust Passports.

Data-flow badges

Every tool is tagged with exactly where its data goes — select a tool in the Workbench and expand Data handling above its content for full details. This works with mouse, keyboard and touch. See the Methodology page for which specific tools fall into each category:

• Local-only — the analysis runs entirely in your browser; nothing is uploaded.
• Local · optional summaries — source files are analyzed in your browser. Improvement summaries and training each require separate permission and start off.
• AutoGovern-hosted — content is processed on our server. Each tool explains whether it also saves a record.
• Third-party AI processing (Governance Copilot, Governance Agent, Policy Compiler) — your question, assessment context or policy wording is sent to a configured LLM provider to generate a response. Organizations’ monthly board report summaries are drafted the same way from the report’s counts and the names of systems with upcoming deadlines, and so is the summary in a governance brief a member has turned on.
• Air-gapped / CLI-only — applies to the downloadable Aegis and Pulse scanners (see §17).

Organization accounts

Every free tool works with zero sign-up. Creating an account at /account is entirely optional, for teams who want:

• A shared organization that multiple people can belong to
• Member roles (owner, admin, member, viewer)
• An audit log of sign-ins and account activity

To create one:

1. Go to /account and choose "Create Account".
2. Enter an organization name, your email, and a password (at least 10 characters).
3. You're automatically made the organization's owner.

Sign in any time afterward from the same page, or from the "Workspace" link in the header.

As an owner or admin, you can invite teammates from the same page:

1. Enter their email and a role (member, admin, or viewer) and send.
2. They get a link to accept — if they already have an autogovern.io account they just sign in; if not, they set a password to create one.

What each role can do on the account page:

• Owner and admin: manage members, issue and revoke Governance API keys, issue, rotate, suspend and revoke agent keys, and mark board reports reviewed.
• Member: add, edit and archive inventory assets and relationships, generate board reports, and compute effectiveness scores.
• Viewer: see everything, change nothing. A refused change says why.

If you belong to more than one organization, a switcher appears automatically so you can move between them.

Anything you save while signed in — a registered AI system, a saved assessment, a Compliance Crosswalk report, or a Vendor Answer Pack — is tagged to your organization and listed on the account page.

Organization work requires membership; anonymous work requires its owning browser. Record IDs do not grant access. Cloud auto-save and optional improvement/training summaries start off and have separate controls. Use Save privately to keep an assessment, then create an expiring read-only summary link in the Dossier only if you want to share. You can revoke that link or delete the source assessment there. Older anonymous records without verifiable ownership are locked.

Owners and admins can change any other member's role or remove them from the member list. Only an owner can promote someone to owner, change another owner's role, or remove an owner — an organization always keeps at least one owner.

Anyone can leave an organization themselves at any time.

Governance automation

Open Workspace → Governance automation (/account#account-automation). The platform's agents check your registered systems every six hours: Autopilot looks for governance gaps, and the Remediation Center looks for risky agent and deployment patterns.

1. Automation level: 1 means recommend only (the default); 2 or 3 let the agents apply a small set of reversible fixes on their own after checking your active policies. Only owners and admins change it. AutoGovern's platform switch caps your level, and while it is off nothing is applied automatically.
2. Recommendations and actions: owners and admins approve, reject, or roll back what the agents raised about your systems, assets and agents. Every decision goes to your audit log.
3. Your Control Plane policies: describe a rule, choose Compile, check the fields, and save it as a draft. Submit it for review; a different owner or admin approves it; then activate it. Active policies govern your registered agents' runtime checks and the automation above.

Your governance brief: in the same section, choose daily or weekly (Mondays) and Save to get an email from the agents about this organization: reassessment alerts, what was recommended or applied automatically, regulatory changes that may affect your systems, deadlines in the next 30 days, incidents and score changes. It is off until you turn it on, it is only for you, and it only arrives when something changed or is urgent. Preview shows what it would say now. A short summary at the top may be drafted by a language model from the listed facts; one whose numbers don't match them is dropped. Every email has a one-click way to stop it.

Evidence coverage also counts Controls Register tests whose latest run passed and was signed off by someone other than the person who ran it, labelled as a reviewed test. Trust Passport levels still use runtime evidence only.

Financial Services

Open Solutions → Financial Services (/finance). The hub is the entry point to the standalone Finance section: the free Finance AI Guide, the Finance AI Vision Board, the Financial Services AI Architecture report, the Finance Governance workspace, the finance Workbench tools (Model Risk Management, Fair Lending, FS AI RMF & Resilience), the finance policy templates and a dated list of regulatory milestones. The regimes table labels each instrument as binding, supervisory guidance, voluntary or pending, and names withdrawn guidance so it is never cited as live.

Free Finance AI Guide

Open Solutions → Finance AI Guide (/finance-ai-governance). The full public guide is free to read without signing in. It covers model inventory and tiering, data lineage and consumer-report use, vendor and DORA third-party review, independent validation with fairness and explainability testing, controlled release with adverse-action and conduct safeguards, and monitoring and incident response.

1. Select Start the how-tos and follow the six steps. Each identifies an owner and a deliverable.
2. Read Worked examples for a credit decisioning model, insurance underwriting with external data and a customer-facing generative assistant.
3. Open Free templates, expand any template, or select Download all templates (.md) to get the model record, validation report, fair-lending review and third-party register entry. No account or email is required.
4. Use Questions examiners ask for SR 26-2 scope, disparate impact after July 2026, EU AI Act timing, DORA and adverse-action reasons, and Official sources for the linked primary guidance.

All learning content is server-rendered and can be read without JavaScript. The optional organization workspace at /finance-governance requires sign-in for private records. Public learning does not expose saved assessments, customer data or dossiers.

Finance AI Vision Board

Open Solutions → Finance AI Vision or the vision board. Explore eight opportunity cards across three adoption horizons — servicing assistants, onboarding operations and compliance assistance first; credit decisioning, fraud and AML, and underwriting once validated; algorithmic execution and advice with evidence. Expand a card to see data, human authority, controls, accountable teams, evidence, monitoring and fallback. Continue to the integration path, five approval gates, suggested roadmap and success measures. Download detailed board (.md) includes all expanded details without an account. These are proposed adoption ideas, not live services or readiness ratings.

Free Financial Services AI Architecture Report

Open Solutions → Financial Services AI Architecture or /finance-ai-architecture. Use the contents to navigate 18 sections covering business modules, the logical architecture with its flowchart, trust zones, data lineage, integration, decision workflows, AI action controls, model risk management, fair lending and explainability, third parties and DORA, monitoring, incident response, the risk register and delivery planning. Choose Download complete report (.md) to save the same report without an account. The report describes a proposed design and separates it from AutoGovern's current metadata capabilities.

Finance Governance

Open Solutions → Finance Governance (/finance-governance), or Workbench → Compliance Dossier → Finance Governance. Sign in to an organization account, then select your organization. This release manages governance metadata; it does not connect to decisioning systems, validate models or monitor live decisions.

1. Select New use case. Describe the decision, entity type, use-case pack, decision impact, most sensitive data, model type, jurisdiction, US states with specific rules, and the business, model and independent validation owners. Optionally link an active AI Registry system. Confirm that the fields contain no customer data and save.
2. Read Overview for applicability triage — SR 26-2, FS AI RMF, ECOA/Regulation B, FCRA, NAIC and state insurance rules, SEC/FINRA, the EU AI Act, DORA, GDPR Article 22, PRA SS1/23, the FCA Consumer Duty, OSFI E-23 and MAS — with official sources. Unknown, not established or excluded from guidance is not an exemption.
3. Under Model inventory, add each model or AI system with impact and exposure; the tier (high, medium, low) is calculated from their product. Record Validations against a model: basis, conceptual soundness, outcomes, implementation, fairness, explainability, findings and conclusion. High-tier models stay flagged until a current independent conclusion exists; first-line self-tests do not count.
4. Use Fair lending & consumer for adverse-action reason codes, the disparate-impact result and the jurisdiction it supports, proxy review and the foreseeable-harm assessment. Verified results need evidence references.
5. Add Third parties & ICT entries with criticality, contract status, exit plan, concentration and validation transparency, then map Data flows with data class, purpose, permissible-purpose rationale and an optional third-party link. Critical providers without an executed contract or exit plan remain visible as gaps.
6. Use Risk register, Controls & evidence (12 starter control groups; not an examination programme), Cases, Releases, Monitoring plans, Manual observations and Changes as in the guide. Live monitoring always reads not connected.
7. Open AI assurance for review domains, evidence gaps, use-case pack hazards (Draft risk opens a prefilled risk) and immutable domain reviews. Have a different organization admin/owner from the creator and latest editor record domain reviews and, under Review & export, the independent dossier review. Edits invalidate reviews; expired evidence expires the current state.
8. Select Download current dossier (JSON) or Download readable report (.md). Exports include metadata, tiers, gaps, review state and an integrity checksum. Review and checksum are not certification. There is no public sharing link.

Healthcare AI Vision Board

Open Solutions → Healthcare AI Vision or the vision board. Explore eight opportunity cards across three adoption horizons. Expand a card to see data, controls, accountable teams, evidence, monitoring and fallback. Continue to the integration path, five approval gates, suggested roadmap and success measures. Download detailed board (.md) includes all expanded details without an account. These are proposed adoption ideas, not live clinical services.

Free EHR Architecture Report

Open Solutions → EHR Architecture or /ehr-architecture. Use the contents to navigate 18 sections covering clinical modules, data, interoperability, deployment, AI controls, governance, monitoring and recovery. The architecture diagram summarizes the system boundaries. Choose Download complete report (.md) to save the same detailed report without an account. The report describes a proposed EHR design and separates it from AutoGovern's current metadata capabilities.

Free Healthcare AI Guide

Open Solutions → Healthcare AI Guide (/healthcare-ai-governance). The full public guide is free to read without signing in. It covers healthcare AI governance, HIPAA/PHI/PII, vendor and BAA review, clinical evaluation, bias, oversight, monitoring and incident response.

1. Select Start the how-tos and follow the six steps. Each identifies an owner and a deliverable.
2. Read Worked examples for a clinical scribe, patient chatbot and predictive clinical model.
3. Open Free templates, expand any template, or select Download all templates (.md) to get the use-case card, vendor review and monitoring plan. The download is free and requires no account or email.
4. Use Official sources for the linked HHS, FDA, ONC, FTC and NIST guidance.

All learning content is server-rendered and can be read without JavaScript. The optional organization workspace at /healthcare requires sign-in for private records. Public learning does not expose saved assessments, patient data or customer dossiers.

Healthcare Governance

Open Solutions → Healthcare Governance (/healthcare), or Workbench → Compliance Dossier → Healthcare Governance. Sign in to an organization account, then select your organization. This first release manages governance metadata; it does not connect to patient systems.

1. Select New workflow. Enter its scope, entity role, data classification, purpose, jurisdiction, clinical impact and accountable team. Optionally link an active AI Registry system in this organization. Confirm that the fields contain no patient data and save.
2. Read Overview for applicability triage and official guidance. Unknown or not established is not an exemption. A reviewer must confirm actual scope.
3. Under Vendors & BAAs, add the exact service, BAA requirement/status, agreement reference, review date and rationale. Then use Data flows to record source, destination, purpose, retention reference and an optional vendor link. Use references, not document contents.
4. Use Risk register for threats, likelihood, highest impact, owner and treatment deadline. The 1–5 likelihood × 1–5 impact score is triage; 15 or greater is displayed as high. Closure requires a verification reference.
5. Use Controls & evidence to record tests against 12 starter Security Rule groups. These are not a complete HIPAA audit. A result other than not tested needs a date and evidence reference. Expired review dates remove tests from current coverage; not applicable does not count as a pass.
6. Use Cases for a reference in your authorized case system, team, status and reviewed deadline. Do not copy incident or patient details. This release does not calculate legal notification deadlines.
7. Ask a different organization admin or owner from the workflow creator and latest editor to open Review & export, record the rationale and choose a future review date. Edits invalidate review; expired evidence or review dates expire the current status. Review preserves known gaps and is not certification.
8. Select Download current dossier (JSON). The authenticated download includes its current review state, scope and checksum. Store it in approved storage; no public link or invitation is created.

Owners, admins and members may edit. Viewers may read and export. Use Reload if someone else has changed the workflow. Archive removes a record from active metrics and exports but retains it in workspace history; it is not deletion. Update linked flows before archiving a vendor. Activity shows the latest 100 events. There are no background alerts, EHR connectors, PHI detection or clinical validation in this release.

Discovery & Integrations

Open Platform → Discovery & Integrations, or use the link in Workbench → Shadow AI Detection. This is a live, read-only GitHub connector. Start with a privately saved assessment.

1. Choose a saved assessment or expand Create a private assessment, name your project, and confirm the private save. Enter a GitHub URL or owner/repository. The new assessment starts without a risk score.
2. For a private repository, provide a fine-grained token limited to that repository, with Contents: read and, if collecting CI, Actions: read. Public repositories may work without a token. Tokens are used for this scan and are not stored.
3. Optionally select Include GitHub Actions run metadata, confirm the read and private metadata storage, then choose Scan repository.
4. Inspect Scan coverage and each item's Inspect collected metadata. Select the items you want, confirm review, and choose Import selected evidence.
5. Choose Open evidence rooms. In a room for that assessment, select the GitHub discovery evidence source, inspect and approve it, publish a revision, and create a customer link.

The scan captures one default-branch commit and reads three root manifests: package.json, requirements.txt and pyproject.toml. It recognizes selected AI dependencies; Python matching is heuristic. Optional CI collection includes up to 20 workflow runs for the same commit, without logs or artifacts. Missing matches do not mean no AI is used. Monorepos, cloud deployments, runtime use and transitive dependencies are outside this connector's coverage.

Scans and imported selections are private. Each imported selection is fixed; run another scan to update it. A collected snapshot does not automatically recheck GitHub, register a system, change a risk score, or certify compliance. Evidence room freshness describes the stored snapshot and its recheck interval, not a live check of the repository. One-time scans do not create recurring connections.

Continuous GitHub connections

No registered system? While signed in to the same organization, open Workbench → AI Registry, enter the system name, type, lifecycle stage, and purpose, then choose Register system. Return to Discovery and click Refresh sync status. Account displays your systems; registration happens in AI Registry. One-time scans do not require system registration.

On the same Discovery & Integrations page, organization owners/admins can choose a registered AI system and configure hourly, six-hourly or daily reads. Explicitly authorize recurring access, encrypted token storage if supplied, and inventory mapping. Public repositories need no stored token; private recurring access requires deployment credential encryption to be configured.

Use Refresh sync status to inspect last success, next attempt, errors and in-app alerts. Sync now queues a read; Pause, Resume, Revoke and erase token, and credential rotation manage its lifecycle. Provider errors back off; repeated failures pause the connection. Acknowledge alerts separately from approving evidence.

Dependency declarations map to stable tool records in the selected system's inventory. Repeated scans do not duplicate records. Removed declarations remain as history; unreadable coverage is unknown. The connector does not establish runtime use.

Choose Refresh scan history to review newly collected evidence. Source changes mark older room evidence changed; unhealthy connections make it unavailable and overdue sync makes it expired. Import the new scan and replace the old draft item before reapproving and publishing. Background collection never changes already disclosed content or auto-approves a customer review.

Customer evidence rooms

Open Platform → Evidence Rooms, or choose Customer evidence room in the Compliance Dossier. Start with a privately saved assessment.

1. Create a room with a customer-facing title and a clear scope, including the system version and deployment covered.
2. Select assessment results, linked vendor answers, or linked signed evidence packages. Add your explanation or answer to a customer question.
3. Inspect the exact content and approve each item for disclosure. Editing or refreshing an item clears its approval.
4. Publish the approved evidence as a fixed revision. Drafts remain private to your organization or owning browser; organization viewers cannot edit them.
5. Create a customer link, selecting exactly which items it includes and when it expires (1–30 days). Optionally allow evidence requests and review decisions. Copy and send the link yourself; the platform sends no email.
6. Answer customer requests in the room. Customers can resolve answered requests and record their review outcome. Each link sees only its own requests and replies.

Customer links stay on their original revision. Publish a new revision and create a new link to share updated evidence. Source changes and expired recheck intervals are flagged when the room is loaded; new source content is never substituted into an approved snapshot. Refresh the page to see new responses or status changes.

Links are bearer access: anyone holding a link can use its permissions until expiry or revocation. A recipient label does not verify identity. Revoke a link under Recipients & review status. Deleting the source assessment blocks room access, but retained snapshots and previously downloaded copies are not erased by that action.

Customers can download their scoped evidence as JSON or a Markdown report. The downloadable Node.js checksum verifier checks the JSON packet's integrity; it does not verify the issuer or certify compliance. Approval and review outcomes are records of this review, not independent assurance.

AI Inventory & relationships

To register an AI system, open Workbench → AI Registry while signed in to your organization. Enter its name, kind, lifecycle stage, and purpose, then click Register system. Use this record for continuous GitHub connections and linked assessments.

While you are signed in, the AI Registry lists every system in the organizations you belong to, plus any this browser registered while signed out. Each row names its organization (or This browser); rows in an organization where you are a viewer say (view only), and changes on their pages are refused.

From the account page, signed-in members can register the assets their systems actually use — models, agents, prompts, tools, MCP servers, datasets, vendors, business processes, and deployments — then link a system to each one it depends on.

This lets you trace any registered system to its full dependency graph — its models, tools, prompts, and datasets — from one place, rather than keeping that knowledge in someone's head.

Registering an asset with a name that closely matches an existing one of the same kind is flagged (not blocked) so you can catch accidental duplicates — the same check applies to systems in the AI Registry.

A relationship isn't limited to system→asset — either side of the "Link" form can be a system or an asset.

So you can also link two assets directly (an agent that delegates to another agent, a tool that's provided by a vendor, and so on), building out a fuller dependency graph than just what each system touches.

Systems now support the full lifecycle: proposed → experiment → design → development → testing → validation → awaiting approval → approved → production → suspended → retired.

Any system registered without an owner email is flagged with a warning on the account page so it doesn't get lost.

The Workbench's background reassessment process — which already re-checked regulatory obligations and evidence coverage on a schedule — now also watches for drift in a system's own details: vendor, geography, autonomy, jurisdictions, lifecycle stage.

It also watches the assets linked to it — a model's version changing, a tool or dataset's configuration changing. Anything it finds shows up in that system's reassessment history in the AI Registry.

Owners and admins can export their organization's entire inventory (every system, asset, and relationship) as a single file from the account page, and import one back in — including into a different organization, which always creates fresh records rather than overwriting anything.

An agent or MCP server asset can also be turned into a real, callable identity that authenticates to the Agent Control Plane — see §7 below.

Agent & MCP Registry

Any AI Inventory asset of kind agent or MCP server can be activated for the Control Plane — turning it from a passive description into a verifiable, revocable identity with its own API key. This reuses your existing inventory instead of asking you to maintain a second, separate registry.

From the account page's AI Inventory list, next to any agent/mcp_server row:

1. Click Activate for Control Plane. You get an API key, shown exactly once — copy it immediately, since it can't be retrieved again afterward.
2. Have that agent (or whatever calls the Control Plane on its behalf) send the key as Authorization: Bearer <key> on every POST /api/control-plane/evaluate call.
3. The action is logged under that asset's real registered name — whatever agent value the request body claims is ignored once a valid key is present, so one agent can't spoof another's identity.

From the same account page list, at any time:

• Rotate key — issues a brand-new key and immediately invalidates the old one.
• Suspend — the next call using that key is denied before policy evaluation even runs, the per-agent equivalent of the fleet-wide kill switch in the Agent Control Plane Workbench tool. Reversible.
• Reactivate — resumes normal evaluation for a suspended key.
• Revoke — terminal. The key can never be reactivated; activate a fresh key if you need to reuse that identity later.
• View ledger — that identity's own slice of the tamper-evident ledger, scoped to your organization only. It never appears in the public anonymous demo ledger, and no other organization can see or manage your agent, even if they somehow guessed its ID.

Governance API

Once you've activated an agent (§7), you can call the same Control Plane your own Workbench uses from any external system — your own agent framework, an MCP server, a CI pipeline — for free, with no separate developer signup or sales call. The full reference lives at /governance-api; this is the short version.

• Send Authorization: Bearer <key> plus a JSON action (at minimum a tool and an action_type) to POST /api/control-plane/evaluate.
• You get back a decision — allow, review, or deny — plus the risk score, matched controls, and reasons behind it, the same shape the Workbench itself uses.
• A review decision can be resolved later via POST /api/control-plane/action/:id/decision, using the ledger id from the response.

Two free, dependency-free SDKs are downloadable directly from the docs page — a single file each, no package manager install:

• Node.js — autogovern.js (uses the native fetch, no dependencies)
• Python — autogovern.py (uses only the standard library's urllib)

Both expose an evaluate() call and an assertAllowed() / assert_allowed() convenience method that raises unless the decision is a clean allow.

Authenticated calls get their own rate limit — 300 requests/minute per agent key — kept entirely separate from the 60-requests/minute-per-IP limit on the free anonymous Workbench demo, so real production traffic isn't throttled at demo-tier levels.

Controls Register & evidence packages

Open any system in the Workbench's AI Registry and scroll to Controls Register to define what should be tested, how often, and who gets escalated to if it fails:

• Title
• Category — Model / LLM, Agentic, Data, or Operational
• Test cadence, in days
• An optional escalation email

Run the relevant Workbench tool yourself and log what it found with one click — Log pass or Log fail:

• The Fairness Scanner for a bias control
• Drift & Data Quality for a drift control
• LLM Safety Check for a prompt-injection/jailbreak control
• ...and so on for the other control categories

Every run is kept. Each result you log is stored as a test run with its history, not overwritten. Open Log a test run with details on a control to record:

• The score, and where it came from: by hand, or the latest result of the Fairness Scanner, Drift & Data Quality, the Document Gap Scanner or the LLM Safety Check in this browser. Only a numeric summary is sent.
• The dataset's name, row count and SHA-256 of the original file, computed in your browser so the data never leaves it. Anyone can reproduce it with shasum -a 256 file.csv.
• The model or configuration version tested, conditions and evidence. The system's version, vendor and linked asset versions are recorded automatically.

Give a control a threshold and say whether a run passes when the score is at least or at most that value. A logged score then decides the result, and a hand-entered result that contradicts it is refused. Run history shows each run against the one before it (regression, recovery, score change, dataset or version changed). Someone other than the person who logged a run can sign it off; on a system registered while signed out, a sign-off is recorded as self-reviewed. The signed evidence package now carries the run history, not just the latest result.

A control that's logged as failing, or that's gone past its own test cadence without a fresh result (overdue), elevates that system's residual risk.

This is shown as a banner at the top of the system record, recomputed automatically every time the background reassessment daemon runs (every 6 hours) — not just when you happen to look.

A failing control also raises the reassessment alert level to critical.

If the system has an owner email or the control has an escalation contact, this sends a one-time email the moment it newly becomes critical — it won't re-send on every run while the issue is still open.

Click Generate evidence package to compile a system's obligations, evidence coverage, controls register, and recent reassessment history into one signed bundle.

It's cryptographically signed with the same key that signs Trust Passports, so anyone can fetch it back by its link and see it re-verify the signature — not just trust that the JSON hasn't been edited.

Next to it, Generate AI-BOM compiles the same system's full dependency graph — every model, dataset, prompt, tool, MCP server, and vendor linked to it in your AI Inventory, each with its version, owner, and status.

It also includes evaluation results and security findings, signed the same way.

If a prior AI-BOM exists for the system, the result also shows what changed since then (assets added or removed, versions bumped) — a shareable, independently verifiable answer to "what does this system actually depend on?"

Impact assessments (ISO/IEC 42005, with an EU FRIA)

In the same system record, the Impact Assessment section (just below the obligation clocks) records who the system affects, how it could harm them, what reduces that harm, who you consulted, and when it has to be looked at again. It follows the structure of ISO/IEC 42005:2025.

• Start impact assessment creates a draft pre-filled from the system's intake answers: suggested affected groups and impacts. Nothing is rated for you.
• Rate each impact's severity and likelihood before and after mitigation. The level appears as you choose; a Severe impact never rates below High. This matrix is AutoGovern's default, and every export says so, because ISO/IEC 42005 does not prescribe one.
• The Completeness list shows what is still missing: context, foreseeable misuse, data, affected groups, impacts, mitigations and residual ratings, benefits, consultation (or why none was held), human oversight, complaints and redress, what happens if a harm occurs, and a review date.
• Section 11 answers whether an EU AI Act Article 27 fundamental rights impact assessment is required for you as a deployer (public bodies and public-service providers, credit scoring, and life or health insurance pricing), and EU FRIA (.docx) exports it in the Article 27 structure.

Approval. For a system inside an organization, an owner or admin who did not start or last edit the draft approves it, and the record shows the approval as independent. A system registered while signed out can only be approved as self-attested, and the record and exports say so. Approved versions can't be edited; Start reassessment creates the next version from the approved one.

Review. An approved assessment is flagged as due for review when its date arrives or when something you ticked in its review plan changes: vendor, a linked model, prompt, tool or dataset, purpose, who it affects, where it is used, autonomy, risk tier, lifecycle stage, or a critical or major incident. Systems that affect people with no approved assessment are flagged in the registry too. These are warnings; nothing is blocked. The signed evidence package includes the approved assessment's summary and a fingerprint of its content.

Incidents, regulator notifications and corrective actions

In the same system record, Internal Incidents tracks what actually went wrong with this system (distinct from Incident Intelligence, which matches other organisations' public failures to your profile). Log an incident, move it through its stages, and open Manage incident to record the owner, when you became aware, who was affected, whether it was a near miss, a serious incident under the EU AI Act or a personal data breach, and the root cause and lessons learned.

• Regulator notification clocks appear for systems used in the EU or UK: EU AI Act Art. 73 for providers (15 days, 10 if someone died, 2 for critical-infrastructure disruption or a widespread infringement), Art. 26(5) for deployers (inform the provider immediately), and GDPR Art. 33 (72 hours) and Art. 34. Record each as notified (when, authority, reference, initial or final report) or not required, with the reason. AI Act clocks show the date the duty applies from and are not overdue before it.
• Evidence is recorded as references to where it is kept; nothing is uploaded. Every change goes into the incident's history.
• Corrective actions come from an incident, a failing control, an auditor finding, or are added by hand. Each has an owner and a due date, and moves from open to in progress to fix submitted to verified. In an organization, the person who submitted the fix and the action's owner cannot verify it; on a system registered while signed out, verification is recorded as self-verified.
• Closing an incident needs a root cause, lessons learned, every in-force notification resolved, and every linked corrective action verified or cancelled. The incident lists exactly what is missing.

The AI Registry list flags overdue notifications and overdue corrective actions (warnings only), and the signed evidence package summarises incidents, notifications and corrective actions.

Complaints, appeals and human overrides

The Complaints, Appeals and Overrides section of a system records what people affected by it raise (complaints, appeals against a decision, feedback) and notable human overrides, with the outcome, whether a person reviewed the decision, whether the AI-assisted decision was changed, and the redress given. It is governance metadata only: record your own case reference, never a person's name, contact details or ID numbers; each new case asks you to confirm that, and text that looks like an email address or a phone or ID number is refused.

• Each case has a response deadline (30 days by default), can be acknowledged, resolved, withdrawn or reopened, and keeps its history. Create corrective action turns it into an owned, dated fix.
• Human oversight by month records decisions made, reviewed and changed by a person, giving real review and override rates.
• The section flags patterns (repeated grounds or groups, an upheld discrimination complaint, a high uphold rate, an override spike; AutoGovern defaults, stated on the page). An approved impact assessment with the complaints trigger falls due for review when one appears, the AI Registry list flags it, and the signed evidence package carries the totals and rates, never case text.

Suppliers

The Suppliers section of a system records the third parties it depends on: who provides the model, hosting, data, tools or services, how critical each one is to this system, what data is shared with it, the fallback if it fails, and the exact model or service used. A supplier belongs to your workspace (your organisation, or this browser if you are signed out), so its contract, data processing agreement, reviews and evidence are recorded once and count for every system that uses it. Record references, never the documents themselves.

• Pick a supplier already in the register, or add a new one. Supplier terms records the contract, data processing agreement, whether it may train on your data, audit rights, availability and incident-notice commitments. Record review sets the next review date from the supplier's criticality (6 months if critical to any system, 12 if important, 24 if standard; AutoGovern defaults, shown on the page).
• Add evidence records a SOC 2 report, ISO/IEC 27001 or 42001 certificate, penetration test and so on with its expiry date; without one, the usual validity for its kind applies. Expired evidence is flagged, and a critical supplier needs current independent assurance.
• The section flags gaps (no executed contract, personal data shared without a data processing agreement, a supplier that may train on your data, a review overdue, a critical dependency with no fallback) and concentration: one supplier serving half or more of your systems, or critical to two or more. Create corrective action turns a gap into an owned, dated fix.
• An approved impact assessment falls due for review when a supplier is added or removed, or the model used from it changes. The AI Registry list warns about supplier gaps and concentration without blocking anything, and the signed evidence package carries each supplier's status.

Change gate

The Change Gate section, first on a system's page, shows whether the system is ready for production (go or no-go). Each check is calculated from the system's records: an approved promotion change request, the impact assessment, controls, incidents, corrective actions, risk acceptances and exceptions, suppliers, generative AI risks, the Trust Passport, and re-tests after changes. Checks that don't apply say so.

• Warnings only, unless your organisation chooses otherwise. By default nothing is refused. An organisation owner or admin can open Gate policy, choose which checks are required and switch on enforcement. Then promotion to production is refused while a required check fails, and an owner or admin can override it with a reason that is logged. Signed-out workspaces are always warnings only.
• Change stage moves the system between lifecycle stages; moving into production goes through the gate.
• Change requests record what is changing, the risk and the rollback plan. An owner or admin other than the requester approves it. After a model, prompt, data or vendor change is implemented, the gate expects every control to be re-tested (and, for generative AI, a new Red Team run); after a purpose or scope change, a newly approved impact assessment.

Generative AI risks (NIST AI 600-1)

The Generative AI risks section of a system lists the 12 risks in NIST's Generative AI Profile (NIST AI 600-1): CBRN information, confabulation, dangerous or hateful content, data privacy, environmental impacts, harmful bias, human-AI configuration, information integrity, information security, intellectual property, obscene or abusive content, and the value chain. For each one you decide whether it applies (if not, say why), write how it is handled, and tick the controls from the system's Controls Register that address it.

• The coverage level is calculated, not chosen: Tested when a linked control was tested within its schedule (or, for information security, a Red Team run in the last 180 days got real answers from the endpoint); Controls in place for linked controls or records such as the supplier register; Documented for written measures or a harm rated in the approved impact assessment. A failing test shows as failing.
• Open incidents of a matching type, failing or overdue controls, successful Red Team attacks and supplier gaps are listed under each risk.
• To count a Red Team run, pick the system in the Red Team tool before running it, or paste an earlier run's link under Link an earlier run. The profile shows only dates and counts, never the endpoint.
• The AI Registry list warns about generative AI systems with risks not yet decided, or an applicable risk with no coverage or a failing test, without blocking anything. The signed evidence package carries the profile.

US notices & rights requests

Open Workbench → AI Registry, open a system and scroll to Notice & Appeal Kit, just below Complaints, Appeals and Overrides. It covers the US laws on automated decisions: Colorado (SB 26-189), the California CCPA rules on automated decision-making (they apply from 1 January 2027), Illinois HB 3773 and New York City Local Law 144.

• Set it up once: tick the states the system is used in, your role and the decisions it makes, then save. The kit lists the notices you owe and what each must say.
• Your wording: fill in the gaps the kit marks, then copy a notice or download it as Markdown or Word.
• Log notice sent when you send one. We keep a fingerprint of the text and the date, never who it went to.
• Rights requests: record each access, appeal, opt-out or data request. The kit works out its legal deadlines and lets you Acknowledge, Extend (California, once) or Complete it. Where the law can be read two ways it shows the legal deadline and an earlier aim for date; a request is late only after the legal deadline. A request entered by mistake can be closed with Recorded in error: it is kept, and left out of every count.

A request past its legal deadline, or not acknowledged in time, raises a warning in the AI Registry list. Colorado dates that come only from draft rules are labelled and never counted as overdue. The signed evidence package carries the counts, not the wording. Record governance details only: no names, contact details or ID numbers of the people affected. This is an educational aid, not legal advice.

AI content marking

Open Workbench → Specialist tools → AI Content Marking. You need to be signed in as an owner, admin or member of an organization.

• Choose a JPEG or PNG of up to 12 MB. A signed content credential (C2PA) is always added. The invisible watermark, the second layer the EU Code of Practice asks for, is on by default; it is skipped for images under 256 × 256 pixels or over the size limit, for very busy images, and for files whose existing AI label it would remove, and the result tells you when. You can add a visible AI badge and China's metadata label.
• Choose Mark this image and download the result. The image is marked in memory and never stored; we keep only a record of the mark (the file's fingerprint, the layers, who marked it).
• Anyone can check a file at Content Credentials, with no account. It says whether the file carries our credential or watermark and whether the file changed after marking. It also shows the name of the organization that marked it and the date, so anyone with the file can see who marked it.

Other C2PA checkers will show our credentials as "valid, not trusted", because they are signed by AutoGovern's own root certificate rather than one on their trust list. You can download it from /api/marks/root.pem. The watermark survives recompression and resizing, but not cropping or rotation.

Free deadline checkers

Open Regulatory Deadlines and pick one of these deadlines: Colorado's AI Act, California's automated decision-making rules, Illinois HB 3773, NYC Local Law 144, Korea's AI Basic Act, China's AI labelling rules or EU AI Act Article 50 marking. Its page has a Does this apply to you? section.

Answer a few questions to see whether the deadline applies, why, and what you owe by when. Nothing is stored and no account is needed: your answers stay in the page address, so you can bookmark or share the result. Each checker shows the date its research was done; our integrity agent tells us when that date is more than 120 days old, and the page says so too.

Trust Passport

Once you've saved a Compliance Dossier, you can mint a Trust Passport — a cryptographically signed, publicly verifiable summary of your system's risk tier, readiness score, and governance checks.

Anyone with the link (or the embeddable badge) can verify it was genuinely issued by autogovern.io and hasn't been altered, without needing an account of their own.

Issue one from the Compliance Dossier panel (optionally add a scope, e.g. "Production deployment, EU region"); verify any passport at /passport/:id.

You can revoke a passport you issued from the same panel — a revoked passport shows a clear "Revoked" banner on its public page rather than looking like a generic verification failure.

Every passport carries a level, computed automatically and never something you can just select:

• Self-Assessed — the default.
• Evidence-Verified — if it's linked to a registered system with real evidence coverage and no failing controls.
• Independently Assured — if an external auditor has signed off on it through the Auditor Portal below.

It also expires (180 days by default) — the public page shows a distinct "Expired" state once that date passes.

From a system's page in the AI Registry, generate an Auditor Portal link to invite an external auditor to review it — no account required on their end, just the link.

They see your evidence coverage and controls-register status (never raw ledger data), can submit findings, and you can post a management response from the same AI Registry page.

When they're satisfied, they sign off with a name and email (self-asserted, not a verified credential — disclosed as such) to issue a signed, public assurance report at /assurance/:id, which automatically upgrades any linked Trust Passport to Independently Assured.

Public trust page

Give customers and buyers one public page that shows how your organization governs its AI systems, at autogovern.io/trust/<your-address>. It is built from your own records each time it is opened, so it stays current.

Open Account → Workspace → Trust page while signed in. Owners and admins can edit it; other members see it read only.

• Choose an address (3 to 40 lowercase letters, digits and hyphens), the name to show, an optional https website and a short summary.
• Say how you use data: whether you train AI on customer data, your data regions, how long you keep data, and a short statement.
• Switch on the sections you want: AI systems, Trust Passports, control testing, incidents (counts only), suppliers, frameworks, data use and your signed effectiveness attestation, private documents, and model, system and agent cards (both below). Every section is off until you switch it on.
• Tick the systems the page may show (up to 50). Only those systems appear, while they are still active in your organization, with the suppliers they currently depend on and the Trust Passports issued for them as publicly listed.
• Choose Save, then Publish. Nothing is public until you publish. Unpublish takes the page down straight away (copies already open in a browser may be cached for up to five minutes).
• A new page says Not verified by AutoGovern and is kept out of search engines until AutoGovern checks that your organization runs it. Changing its name, address or website clears verification. An old address redirects to the new one and stays reserved to you.

The page never shows owner emails, internal notes, contract references or incident text. Your wording is refused if it looks like it contains an email address or a phone, card or ID number, and system and supplier names, data regions and model versions that look like that are withheld. Names of people cannot be detected automatically, so keep them out. Revoked, expired and unlisted passports never appear. The same data is available as JSON at /api/trust/<your-address>. The page says plainly that the facts come from your records and that AutoGovern does not certify them.

Confirm your domain

Owners and admins can show that they control the domain of the page's website. Open Account → Workspace → Trust page → Confirm your domain.

• Save the page with your website first. The card then shows a TXT record name and value. Add that record at your DNS provider and choose Check now.
• DNS changes can take up to an hour. You can check ten times in ten minutes.
• AutoGovern checks the record again every week. If it is gone, the confirmation is removed; a lookup that cannot finish changes nothing. A new website removes the confirmation too.
• A confirmed domain helps AutoGovern verify your page, but does not verify it on its own. The public page does not show it.

At the top of the workspace, owners and admins also see a Publish your trust page checklist: save the page, turn on at least three sections that have content, publish, confirm your domain, and wait for AutoGovern to verify. A section counts only when it shows something (for example AI systems with at least one system). Without a website DNS can check, the domain step is not needed. The card goes away once AutoGovern has verified your page; Hide hides it in this browser.

Private documents on request

Offer documents buyers ask for, such as a SOC 2 report or a pen test summary, without emailing files. Only titles are public; the files go only to people you approve. Manage them in the Private documents and Access requests cards under Account → Workspace → Trust page (owners and admins; other members read only).

• Switch on Private documents under What the page shows, save and publish the page. Until then buyers cannot see or request anything.
• Upload a PDF (up to 10 MB) with a title buyers will see, or offer a record you already have here: a signed evidence package or AI-BOM (sent as signed JSON) or a vendor answer pack (sent as Markdown). Up to 25 documents and 150 MB of uploads per organization. The uploaded file's own name is not kept.
• Write your NDA under NDA and choose Save NDA. Each change is a new version. Documents marked Buyers must accept the NDA cannot be requested until an NDA is saved.
• A buyer chooses Request access on your trust page, signs in with a free AutoGovern account, ticks documents, gives a short purpose and accepts the NDA where needed. Before sending, they are told that you will see their name, email address and organization. They can have one pending request with you at a time and can withdraw it.
• Under Access requests you see each requester's name, email address and the organizations their account belongs to, the documents, the purpose and the NDA version accepted. The requester's details are self-declared (accounts are not email-verified) and only owners and admins see them. Approve for the number of days you enter (1 to 90), and nobody can decide their own request, Deny with an optional note, or Revoke access later.
• The buyer downloads from My access on the request page while access lasts. Every download, including checks by your own owners and admins, is listed under Access log.
• If email is set up, your owners and admins get a short notice when a request is waiting (no requester details, at most one every ten minutes). Buyers are not emailed: they see decisions on the request page. Nothing emailed gives access.
• Archive withdraws a document from the page and from every request; an archived PDF's file is removed from the database, while its title and download history are kept. Signed evidence packages and AI-BOMs also have a public verification link by id; the id is never shown on the trust page.

Answering buyer questionnaires

Answer a buyer's security questionnaire from your own records, in the Questionnaires and Answer library cards under Account → Workspace → Trust page (owners and admins; other members can read and export).

• Draft from records writes draft answers from your answer packs, trust page data-use answers, publicly listed passports, AI-BOMs, evidence packages, controls, suppliers, incident counts and approved impact assessments, each with its evidence. Running it again never changes an answer you approved, edited or archived.
• A person approves every answer. Yes or Partly needs evidence; No or Not applicable needs a short explanation.
• Upload the buyer's .xlsx or .csv (up to 5 MB). The question and answer columns are found from the headers, and each question gets the closest answer from your library. Edit, approve or skip each one, or approve the confident suggestions in one step.
• Export writes only approved answers into the buyer's file and leaves everything else as it was, or offer the finished questionnaire on your trust page as a private document that buyers request and download like the others.
• Buyers see links only where they are public (your trust page, a Trust Passport's verification page). Signed AI-BOMs and evidence packages are named, never identified.
• CSA material: if any sheet holds three or more CSA-style question IDs (like A&A-01.1), or the file names CSA, CAIQ, AICM, CCM or STAR anywhere, or a sheet is too large to check, the whole file is handled in IDs-only mode: no question text and no file are kept; select your copy of the same file again to see the text, export or suggest. Shared Assessments SIG files are refused (licence). A buyer's own questionnaire is kept, with its file, until you archive it; archiving deletes the file and the question wording.
• Approving suggestions in bulk only approves those taken unchanged from answers you already approved in the library. Buyers who are approved for an offered questionnaire get a new sheet with the questions and your approved answers, never the uploaded file.

Model, system and agent cards

Publish cards in the shape US federal buyers ask for under OMB Memorandum M-26-04 (Appendix A): what a model, system or agent does and is for, how it was trained or built, its risks and how you reduce them, evaluation results, your acceptable use policy, end-user resources and how to give feedback, and the enhanced items an agency may ask for. Manage them in the Model, system and agent cards and Your cards cards under Account → Workspace → Trust page (owners and admins; other members read only). A card is your organization's own statement: AutoGovern prefills it from your records and you edit it. AutoGovern does not certify it, and every public view says so.

• Create a card: choose the kind (model, system or agent), the system from your AI Registry (required for a system card) and, if you like, an inventory asset (a model, or an agent or MCP server).
• Draft from records fills only empty fields, each with a note saying where it came from: the system's purpose, vendor, version, risk tier and lifecycle stage; the asset's description and version and the names of linked models, agents, tools and MCP servers (never datasets, deployments or vendors); control test runs of the last 365 days by model version; Red Team run counts as security red teaming (never prompts or answers; they do not count as red teaming for bias); the approved impact assessment; generative AI profile coverage; incident counts; your trust page's data-use statement. It never changes a field that has a value.
• Each field explains what to write and which M-26-04 item it answers. A meter shows how many of the 4 minimum (§1.A) and 12 enhanced (§1.B) items the card says something about; item 1.A.ii is the card itself, with the parts such cards often include listed under it. That is not a judgement that the answers are enough.
• To publish, fill in what it is and does, its intended uses and its known risks and limitations; an agent card also needs its https endpoint, protocol binding, A2A protocol version, input and output types and at least one skill. Text that looks like an email address or a phone, card or ID number is refused (except a shared inbox such as feedback@ your domain in How to give feedback), links must be https, and evidence package or AI-BOM ids are refused. Results, links and skills have a row per entry with a box for each part.
• A card is public only when it is published, your trust page is published, and Model, system and agent cards is switched on under What the page shows. Edits stay private until you choose Publish changes. Unpublish or Archive takes it down. Up to 100 cards.
• Buyers open each card at /trust/<your-address>/cards/<card id> and can download it as Markdown (Hugging Face model card layout), as a CycloneDX 1.7 ML-BOM, and for agents as an A2A agent card. The linked system's name is the one at publish time, shown only while your trust page lists that system.
• AutoGovern signs an agent card only when your trust page is verified by AutoGovern at the moment you publish it; otherwise it is published unsigned and every view says why. If the page loses or renews its verification, the signature is withheld until you publish the card again. The card's status link says whether it is current, unsigned or withdrawn. A valid signature shows the agent card is the one you published, unchanged; not that it is true. A copy saved before a card was withdrawn keeps its signature and cannot be revoked; the status link is the check.

Following your vendors' trust pages

When a supplier of yours publishes a trust page on AutoGovern, link it to that supplier in your supplier register, get told when it changes, and copy some of what it says into the supplier record. Open Workbench → AI Registry, open a system, and use Vendor trust page on a supplier under Suppliers; every link is listed under Account → Workspace → Vendors. Owners, admins and members can change links; viewers read only. Organization suppliers only. Imported facts are the vendor's own public statements, verified by AutoGovern or not, and do not replace your own review.

• Connect by the page's address (like acme or /trust/acme): published pages only, and an old address leads to the current page. The page is public, so the vendor is not asked or told.
• Or Create invite link, copy it and send it to the vendor yourself: AutoGovern never emails it. It works once, for 30 days, and a new link replaces it; an unanswered one is marked expired. An owner or admin of the vendor opens it, signs in, chooses their organization, sees exactly what you will see (their published trust page) and your organization's name (not yours), and accepts or declines (both need an owner or admin). After accepting, they see your organization under Buyers following your trust page in Account → Trust page.
• You see only the vendor's public trust page. Whether it is published and verified is always shown as it is now, with Not verified by AutoGovern wherever AutoGovern has not verified it. Completed questionnaires are listed as available on request, with a link to ask; private documents still need a request the vendor approves.
• Import selected adds Trust Passports and published cards as vendor statements, kept apart from evidence you hold: they never replace your own items, never expire and never clear a gap. They are added once each, even after the vendor changes its address. Data regions, and whether the supplier may train on your data, are copied only where your record is empty; a vendor's "never trains on customer data" is shown as a statement to check against your contract, not written as a term. Nothing you entered is overwritten.
• Pages are checked again every 24 hours, or choose Refresh. What changed is listed, and Changed since last import appears until you import again. A page that cannot be read three times in a row stops being checked until you refresh it. Revoke stops following; what you imported stays.

AICM & AIUC-1 mapping

Buyers often ask about the CSA AI Controls Matrix (AICM) and AIUC-1. AutoGovern maps every AICM 1.1 control ID (247) and every AIUC-1 requirement ID (55, Q4-2026 release, 5 of them retired) to its anchor controls in NIST AI RMF, ISO/IEC 42001 Annex A and the EU AI Act, so a control you register and test once counts toward each.

• Public lists, no account: open /standards, then AICM or AIUC-1. Each ID has a one-line topic in our own words, the mapping strength (Full, Partial or Not covered), the anchor controls it maps to and a Disputed marker where our two analysts disagreed. AIUC-1 IDs link to their official pages. The same data is JSON at /api/standards/aicm and /api/standards/aiuc-1.
• Your readiness: open Account → Workspace → Standards readiness while signed in (every member, read only). Choose the standard and, if you like, one system. Each ID shows Ready, Partly ready, Gap, Not covered by our anchors or Retired, and which anchor controls are not met yet (not registered, not tested, failing, or a pass older than the control's evidence freshness window). A control counts when it is registered with the anchor's framework and control id; choose it under Satisfies framework control when adding a control in the Controls Register (the Governance Autopilot's "create missing control" action sets it too).
• Compliance Crosswalk: each row of Workbench → Compliance Crosswalk shows Also covers with the AICM and AIUC-1 IDs mapped to its controls; the .docx and .md exports list them too.
• CSA questionnaires: for an IDs-only CSA questionnaire, answers that cite a controls register for a framework our mapping names for that question get a small preference. It only adds to answers that already match the question's words.

How to read it: the mapping is AutoGovern's own analysis, made conservatively (two analysts mapped each ID independently; where they disagreed the lower strength was kept). It is not an official CSA or AIUC crosswalk. A partial mapping is never shown as Ready, because our anchors cover only part of that ID. Not covered means our AI-governance anchors do not address the ID (most cloud-infrastructure controls), not that your organization fails it. AICM 1.0.x questionnaires use older numbering for some IDs, so questions whose ID means a different control in 1.0.x and 1.1 get no mapping-based hint; IDs are never matched across versions on the ID alone. We re-check both standards by hand (AIUC-1 publishes quarterly; its next release is scheduled for 15 January 2027).

Control and requirement IDs belong to the Cloud Security Alliance (AICM) and the Artificial Intelligence Underwriting Company (AIUC-1). Descriptions and mappings are AutoGovern's own; read the official standard for the requirement text: AICM v1.1, AIUC-1 and the AIUC-1 standard.

AutoGovern Rating & Diligence Reports

Every publicly listed Trust Passport also carries an AutoGovern Rating — a single, comparable A–F score recomputed live from readiness, assurance level, your Control Plane telemetry, and Incident Intelligence exposure, weighted by how much governance quality actually matters for that system's domain (its materiality).

• The full, ranked list — /ratings
• The published methodology behind every number — /standard
• A live-generated aggregate report — /report

The rating panel on your passport page includes a copy-paste embeddable badge.

From a registered system's page in the AI Registry, generate a signed Diligence Report — either an investor report (M&A / portfolio diligence) or an insurer report (underwriting submission). It compiles that system's Rating, AI-BOM, evidence coverage, and incident exposure into one document at /diligence/:id.

A board or investor audience can start at /board instead of the practitioner Workbench.

Aegis & Pulse (free downloads)

• Aegis is a free, standalone security scanner (macOS/Windows/Linux) that finds real misconfigurations on your machine — fully offline, air-gapped by default.
• Pulse is a free command-line tool that re-ranks your existing vulnerability report (or scans a path directly) against live CISA-KEV and EPSS exploit data, so you can tell which CVEs are actually being exploited.

Both are downloadable from the homepage (Aegis, Pulse) with no account required.

The free public reference (no account)

Everything below is open, needs no sign-in, and is maintained by the same agents that run the Workbench. Each page is a permanent URL you can bookmark, cite or send to a colleague.

• 60-second check — six questions, then your risk tier and the deadlines that actually apply. Nothing is stored; the answer lives in the URL, so it is shareable.
• Regulation reference — one page per regulation we track: status, catalogued requirements, the controls that satisfy each one, and the dates attached to it. Every page shows when our agents last read that regulation's own official source, and warns you when it has moved since we wrote the summary.
• Compare regulations — two or three regulations side by side on the same obligation themes. It shows where they overlap, what only one of them demands, and which deadline lands first. Useful when you already comply with one regime and need to know how much of the next one you get for free. An empty cell means we have not catalogued a requirement there, not that the law is silent.
• Regulatory deadlines — every dated obligation we track, filterable, with a calendar feed you can subscribe to so new ones appear on their own. Seven of them have a free "Does this apply to you?" checker.
• Content Credentials — check whether an image carries our signed content credential or invisible watermark, and whether it changed after marking.
• Incident radar — curated AI failure case studies plus a live feed from the news, each curated one with its own page and the control that would have prevented it.
• Show your work — our own agents' record, including the failures: what they changed with no human involved, what our own integrity agent currently says is wrong with us, every prediction this blog has committed to and whether it came true, and the official links we publish that turned out to be dead. It is the standard we argue for on this site, applied to us.
• Benchmark index, Trust Registry, Ratings leaderboard, State of AI Governance and the weekly digest.

These are plain-English summaries written by us, not legal advice. Every page links its official source — read that before you rely on anything.

Private Governance Studio

Open Platform → Private Governance Studio, also available in the mobile Solutions menu. Add UTF-8 text, Markdown, CSV or JSON files, select the evidence to review, and choose Run local checklist. The six-topic checklist identifies phrases to inspect; it does not prove controls work or change platform scores.

Level of detail starts at General overview. Choose Granular for explanations and review history, or Micro granular for exact source passages and line references. Each review retains its own source snapshots. Removing a current source does not remove it from earlier snapshots.

Use Save encrypted project to download an encrypted backup and Open encrypted project to restore it. Use a passphrase of at least 12 characters and keep it safe; AutoGovern cannot recover it. Backups include all sources and the last ten reviews. Export review as Markdown produces an unencrypted report at your selected detail level. Unsaved page contents are lost on reload.

For model assistance, open Use my API key or local model, download the companion, and run node autogovern-studio.cjs using Node.js 22 or newer. Open the printed local address, load your sources or encrypted project, then choose Use my own model. This increment supports Anthropic and local Ollama. Enter an exact model ID, a per-request key for Anthropic, and an output token cap. Review the destination and selected files before sending. Keys are not saved. No automatic retries or provider fallback occur.

For local Ollama, disable cloud features in the Ollama server configuration (OLLAMA_NO_CLOUD=1) and restart it; use an installed local model. Anthropic requests send selected names and text to Anthropic and can incur charges. Output caps are not dollar budgets. Model references are validated, but findings still require human review. No PDF/OCR, background evaluation schedule, release gate, OS credential vault or enterprise runner is included in this increment.

Learn more

• Methodology & Scoring — the exact formulas behind every score and classification.
• Privacy Policy, Data Processing Addendum, Terms of Service.
• Questions: info@autogovern.io.