Shadow AI & asset graph

Shadow AI: the register

• The Shadow AI page has five tabs: Overview, Register, Discover, Catalogue and Policy. Add a tool opens a search of the AI tool catalogue; picking a tool fills in its vendor, category and data-handling facts. If it is not listed, add it by name.
• Each tool has a decision: Needs review, Under review, Approved, Approved with conditions, Blocked or Replaced (blocked, with a safer tool to use instead). Conditions are required for Approved with conditions, a reason for Blocked, and a replacement for Replaced. Approvals get a yearly review date unless you pick one.
• Every tool has a 0–100 risk score with the reasons listed under Why this risk score: the catalogue's baseline for the tool, whether it has been reviewed, the kinds of information that go into it, and whether it is used for decisions about people.
• Give each tool an owner; they are notified. From More actions you can register the tool as an AI system, add it to vendors (optionally logging a vendor risk), or show it on the Governance Map.

Shadow AI: finding tools

• The Discover tab reads an expense or card export, an SSO app report (Okta, Microsoft Entra ID, Google Workspace), dependency files (package.json, requirements.txt, pyproject.toml, .csproj, go.mod, pom.xml, Gemfile, .env.example), an existing spreadsheet of tools, or the project's latest compliance scan. Columns are recognised automatically and the ones used are shown.
• Nothing is added until you have checked the matches. Untick anything that is wrong, then add the rest. Tools already in the register get the new evidence instead of a duplicate. Values in .env files are never read, and rows that do not match an AI tool are not stored.
• Ask your staff creates a link to a form people can open without signing in. They pick the tools they use, say what for, what information goes in and how often, and can stay anonymous. Each answer adds or updates a tool, and you can switch a link off at any time.

Shadow AI: the catalogue and community

• The AI tool catalogue is shared by every workspace. Each entry shows whether the tool trains on customer data, where data is stored, certifications, the data people usually put in, suggested controls, safer alternatives and its sources. Entries marked Verified have been checked against their sources by AIGovernr.
• Share with the community sends a tool that is not listed, or a correction to one that is, for review. Only tool-level facts are sent: never your organisation's name (unless you choose to be credited), how or why you use the tool, who uses it, spend, notes or decisions. Details about data handling, storage or certifications need a source link.
• When an important change is made to a catalogue entry, every register using that tool shows Catalogue updated and its owner is notified, so the decision can be re-checked. Keep my decision records that you have reviewed the change.
• Usage counts on catalogue entries only appear once five or more organisations use a tool, and are shown in bands such as 10+.

Shadow AI: acceptable use policy

• The Policy tab writes an acceptable use policy for AI tools from your decisions: approved tools, tools approved with conditions, tools people must not use (with what to use instead), tools still under review, and rules for every AI tool. Copy it, download it, or save it as a draft in Policies.

Asset Graph

• A canonical list of AI assets, including AI systems, models, datasets, prompts, tools, MCP servers, agents and vendors, and the dependencies between them.
• Impact analysis shows what a change to one asset affects. Reconciliation matches discovery observations to existing assets, and history shows how the inventory has changed.
• Agents governed in Agent & MCP Governance are created here first.

Governance Map

• Governance Map draws the Asset Graph as a map and adds the controls, risks, findings, evidence, approvals, incidents and vendor assessments that sit around each asset. You need permission to view assets; areas your role cannot see are left out and the page says so.
• Around each asset the map draws the records that govern it: the vendor that supplies it, the frameworks it is held to, and its risks, findings, incidents, approval and evidence. A dashed box marks a record that should exist but does not, such as no controls or no approval. A supplier already on the map is linked to the assets it supplies even if nobody drew the line. Use the records switch to show them for every asset, only the selected one, or none.
• Each asset shows a ring and a score. The lens changes what the ring measures: overall health, control pass rate, open risk, evidence currency, approval status or ownership.
• Governance gaps are worked out for every asset, such as no owner, no controls, expired evidence, a stalled approval or an unapproved vendor. The Fix queue groups the same gap across assets so you can see and fix it once. An asset that depends on something with a serious gap is marked as inheriting that exposure.
• Select an asset, or click any record around it, to open its details. The panel opens with a one-line verdict and a scorecard, then a tab for each kind of record: controls, risks, findings, incidents, evidence and approvals. Each row shows its status, owner or assignee and due date, flags what is overdue, and opens the page where you work on it. The first rows are the most urgent.
• Start here ranks the assets that most need attention. Concentration risk lists the assets that many others depend on. Hover over any node for a quick summary.
• Owners can be assigned from the Details panel. Every other fix opens the page where it is resolved.

Work on the map

• The map fills the page. The search box, lens, records, layout, grouping and filters sit along the top of the map, the tools sit on its right, and the details panel docks on the right edge. Drag the panel's edge to resize it, or collapse it to a rail. On a phone, the controls collapse into one View button and the panels open as a sheet from the bottom.
• Drag any node to place it. It stays where you put it, and Undo and Redo step through your last moves. Hold Shift and drag to select several nodes and move them together. Your arrangement is saved in this browser for this project. Reset layout clears it, and Tidy up lays out the nodes you have not pinned. Right-click a node (or press and hold on a touch screen) to focus on its neighbourhood, simulate from it, trace a path to another asset, pin it, hide it, or open it in the Asset Graph.
• Group by owner, lifecycle, vendor, risk tier or layer to draw a box round each group. Collapse a box to a single node to tidy a large map. Zoom out and nodes become compact; zoom in and the top gaps show on the node itself.
• Press Ctrl or ⌘ with K to search every asset and action from one box. Press ? for the full list of keyboard shortcuts.

Cards, owners and live numbers

• Card detail sets how much each asset card says. Compact shows the name and the ring. Standard shows the reading for the chosen lens and the gap count. Rich adds the owner, the EU AI Act tier, a cell of live numbers for each module (controls passing, open risks, open and overdue findings, evidence that is current, expiring or expired, the approval, open incidents, open tasks and open alerts) and a bar showing how many open risks are treated by a passing control with current evidence. Press a cell to open that module in the details panel. The choice is remembered for you, and a saved view or a link carries it too.
• Every number comes from the module it belongs to, so it only shows if your role can see that module. A module you cannot see shows a dashed cell that says so. Tasks only count when they are tied to the asset's AI system, and a policy only counts when it maps to one of the asset's controls.
• The Treated figure in the header is the share of open risks, across the map, that are treated by a passing control with current evidence. Press it to colour the map by treatment. Due shows how many things fall due or expire in the next 30 days (evidence, waivers and supplier dates). Press it, or use Due in 30 days, to highlight them.

Unfold an asset into its records

• Press the plus button under an asset card, double-click it, press E, or choose Show its records from the menu to unfold it. The asset's own risks, controls, evidence, findings, tasks, remediation plans, incidents, waivers, approvals, monitoring, evaluations and policies appear beside it, in columns that follow the chain: risks, then the controls that treat them, then the evidence that proves the controls; findings, then the tasks and plans that fix them; and the rest in a row underneath. Press the button again to fold it. Show the records of every asset is offered while ten or fewer assets are drawn.
• Lines say how records are linked, in words: treated by, proven by, raised against, fixed by, implemented by, waives, governs and tests. A dashed red box is a link that is missing: No control, No evidence or No fix assigned. A small label on a card says what is wrong with it, such as overdue, not proven or expiring. A line that says +N more means the asset has more records of that kind; open it to see the whole list.
• Assets that belong to the same AI system share their risks, controls and evidence. When two such assets are both unfolded, the shared records are drawn once, in the shared lane, with a line to each asset.
• Press a record to see it in the details panel: who owns it, when it is due or expires, everything it is linked to, and what is missing. The panel links to the record in its own module. Risks and approvals open on their own page. Controls, findings, evidence, tasks, incidents and remediation plans open that record. Waivers, monitoring and policies open the module's list, because those pages do not open a single record.
• Records can be dragged and pinned like assets. Which assets are unfolded is part of your arrangement and of a saved view.

Layers and people

• Layers choose which kinds of record the cards show cells for and which records an unfolded asset draws. Start from Risk treatment (risks, controls, evidence), Audit readiness (controls, evidence, frameworks, approvals, policies), Issue tracking (findings and tasks), Operations (incidents, monitoring, evaluations) or Accountability (people), then switch single layers on or off. Layers are kept in the link and in saved views.
• The People lens colours each asset by how much open work its owner is carrying. Group by owner turns the map into one lane per person, each with a header that counts their assets, open risks, findings and tasks, and what is overdue. Assets with no owner are in an Unowned lane that comes first. Press anyone's initials on a card or a record to highlight everything they own, records included, and press again to stop. Names are never put in the address.

Simulate impact

• Simulate impact asks a what-if question about one asset: it fails or is withdrawn, it changes (a new version or drift), it is compromised, or what it relies on. You can start from several assets at once. The run is shown on the map hop by hop; use the player at the bottom to play, pause or step through it.
• Each reached asset has an impact score. The score falls with each hop and with how weakly the link is likely to carry the problem, so a direct dependency scores higher than a distant one. A vendor link inferred by the map counts for less than one drawn by a person. Choose Why is this affected? on any asset to see the path that reached it.
• The result gives how far it reaches, an impact rating (contained, moderate or severe), the approvals and controls to re-check, the owners to tell, and a list of actions. Copy brief turns it into text for a client note.
• A run can also start from a record. On an unfolded control choose What if it fails?, or on a piece of evidence choose What if it expires?. The panel says what is lost along the chain, for example that four risks lose their only control across three systems, or that a control can no longer be proven. It lists the records affected and the best fixes, such as linking a second control or renewing the evidence. The assets that carry the exposed risks then start the usual run, each as strongly as its risk is exposed. It is based on the records that were loaded, and says when an asset has more than that.
• Try mitigations to see what would change if you fixed something first, such as adding an owner or approving a vendor. The panel shows the result before and after, and lists the fixes that reduce the impact most. Compare with another run shows what changed against a run you saved earlier.
• The Whole project tab runs the same scenario from each asset and ranks the assets by how far a failure would reach. It can be cancelled. The Blast radius lens colours the map by the same result.

Impact scores are an estimate from the links on the map. Treat them as a way to compare assets, and check the assumptions listed with each run.

Saved views, saved runs and sharing

• Views saves the lens, layout, grouping, filters and your arrangement under a name. You can mark one as your default and share it with the project. Anyone who can view assets can open a shared view. Only the person who made a view, or someone with full asset access, can change or delete it.
• Save run keeps an impact run with its scenario and results. Open it from the Saved tab and choose Re-run on today's data to see how the impact has changed since.
• The address in your browser holds the lens, layout, grouping and selection, so you can paste it to a colleague. Export saves the map as a PNG. Report opens a printable page for an impact run, which you can print or save as a PDF from the browser.
• From a run, Raise incident or Log a risk from this run opens that page with the title and description filled in from the run. You need access to the incidents or risks area to save them.