AI Command Center — Customer Portal

The AI Command Center is AINOW's governance module for AI. It finds the AI apps, agents, models and integrations in use across your organization, checks them against a set of controls, and produces signed evidence an auditor can verify offline. Open it from Command Center in the portal's left-hand navigation.

Who can use it

The AI Command Center is licensed separately from chat, voice and agents, and it appears as AI Command Center (monthly) on invoices. You can use it once your AINOW contact has made it available to your tenant or, if a provider manages your tenant, once that provider has licensed it for you. You cannot switch it on yourself. Without the licence, Command Center does not appear in the navigation, and a direct link shows a notice naming who to contact.

If it is the only module on your plan, the portal opens on the Command Center and leaves out the pages for chat, voice and agents. Zoe shows a page that points you to the portal.

Administrators hold every role. You can give other members these roles on the Members card in Settings:

AI Command Center viewer can open the pages and read them.
AI Command Center manager can run scans, switch and connect sources, change statuses and owners, notify owners, request evidence, and change settings.
AI Command Center people detail can see who used an asset, when your organization has chosen named visibility.

The pages themselves need the viewer role, so give it to managers and people detail holders as well. See Signing in and access for members and roles.

The pages

Command Center opens on one page with three parts: the overview, Discovery sources and Inventory. Links at the top lead to the rest:

View the controls opens Controls, with each control's result and findings.
Settings holds the options for people, the digest, retention and the inventory. Managers also find Catalogue overrides there.
Evidence is where bundles are requested and downloaded.

The overview shows your Posture score, AI assets, Waiting for review, Sanctioned and Last scan, the totals for Controls passing, Controls failing, Not applicable and Errors, and charts of the score over recent scans, assets by kind and status, and findings by severity. Its counts cover AI assets only.

Connecting sources

A source is somewhere the Command Center reads from. Every source only reads, and each Microsoft row lists the read permissions it uses.

AINOW platform reads the models, integrations, agents and phone lines in your workspace, and their model usage. It needs no setup and is on from the start.
Entra applications reads the enterprise applications in your directory that match known AI apps, with their consent details.
Entra sign-ins reads sign-in activity on those apps. It needs Microsoft Entra ID P1.
Entra agents reads agent identities in your directory. It depends on a Microsoft preview API.
Power Platform agents reads your Copilot Studio agents and their owners.
Microsoft 365 Copilot usage reads Copilot activity. It needs Microsoft 365 Copilot licences.

To turn on the Microsoft sources:

In Discovery sources, ask a Microsoft Entra Global Administrator to select Open admin consent and approve it. The button beside it copies the link so you can send it to them. One consent covers all the Microsoft sources.
Turn on each source you want with its Enabled switch.
Select Test connection on the row. A working source reads Connected.

Power Platform agents takes one more step. After the admin consent, select Connect and sign in with an account that holds the Global, Power Platform or Dynamics 365 administrator role, or the Global reader, AI administrator or AI reader role. Connecting turns the source on, the row shows who connected and when, and the connection renews itself nightly. If it is revoked or expires, select Reconnect. Disconnect removes the connection and turns the source off.

Under Status, a row reads Connected, Consent needed, Licence missing, Setup pending, Throttled, Error or Not probed yet, with a sentence saying what to do next. If it says no directory is bound to your tenant, ask your AINOW contact or your provider. Last scan keeps the outcome of the source's last scan, and the reason if it did not succeed, even after a later test passes.

Scans

A scan runs every night. A manager can also select Run scan, and the results appear within a few minutes. If a scan ran in the last few minutes, wait before starting another.

Each scan updates the inventory and then evaluates the controls. Last scan reads Completed, Completed with issues or Failed. Completed with issues means something did not finish, usually one source. What the other sources collected is kept.

The inventory

Inventory lists every asset with its Name, Kind, Vendor, Status, Owner, People (30d), Events (30d), Last seen and Sources. Search by name, or filter by kind, source and status. Known AI apps are recognized from a catalogue that AINOW maintains, and an asset seen by two sources appears once.

A new asset starts as Discovered, which the overview counts as Waiting for review. A manager records a decision from the menu on the asset's row: Under review, Sanctioned, Tolerated, Unsanctioned or Retired. A status is a record of your decision. It does not block the app or change anything in the system it came from. A Retired asset returns to Discovered if a later scan sees it again.

Select Details on a row to open the asset:

Overview shows its facts, including its Owner, its Risk tier where the catalogue sets one, and its use over the last 30 days.
Sources shows which sources reported it. For Entra applications it also shows the publisher, the type of consent and the scopes granted.
Controls lists the controls it fails in the last scan.
Timeline records status, owner and risk changes, each time someone showed its people, and each time it was seen again.
People appears only with named visibility, for members who hold the people detail role.

Controls and the posture score

Thirteen controls are evaluated at the end of every scan. On Controls, each one shows its severity, its result and how many findings it has. Select Show findings to see the assets involved. The controls check that:

every asset that is not retired has a named owner, including every Copilot Studio agent, and nothing waits in Discovered for more than 30 days
no app from an unverified publisher holds consent for everyone in the directory, and no user has given an app that is not sanctioned access to mail, files, sites or the directory
no agent or phone line has gone 90 days without activity, and none with tools reaching mail, files, HR or finance systems is still waiting for review
no enabled phone line is set to Always approve
no app marked Unsanctioned was used in the last 30 days
every AINOW model call went to a catalogue model or a configured bring-your-own endpoint
PII masking and the input/output scan are on for every AINOW channel, and the audit chain is anchored daily
every sanctioned app has a data-processing agreement and a data residency on file in the catalogue

A control reads Pass, Fail, Not applicable or Error. Not applicable means it cannot be judged yet, for example because its source is off. Error means one of its inputs could not be read.

The Posture score runs from 0 to 100. It is the share of controls that pass, weighted so that higher severities count for more, and only Pass and Fail count. The score can therefore move when you turn on a source: a control that was Not applicable starts to count.

Entra agent identities and external agents report no activity or tool data, so the dormancy and over-broad tools checks do not judge them. Each control also lists the framework references it relates to. These are pointers for your own review. A passing control is not a certification or a statement of compliance.

Owners

When a source reports an owner, as Copilot Studio and Entra agents do, the inventory shows it with from Power Platform or from Entra, named from your synced directory where possible.

To assign an owner, open Details, select Edit owner, enter an email address, a UPN or an Entra object id, and select Save. An owner you assign takes the place of a reported one. Clear owner removes it.

Notify owner emails the owner a request to review the asset, with its status and a link to the Command Center. The email goes out at the next hourly notification run, and a reply goes to the person who sent it. You can notify the owner of an asset once every 7 days, and the page tells you when you can send the next one. The owner needs an address: either the email address you entered, or an entry with an email address in your synced directory.

Seeing who used an asset

By default the Command Center shows how many people used an asset, never who. An administrator or manager can change this under People visibility in Settings.

With Named people selected, members holding the people detail role get a People tab on each asset. It loads only when they select Show people, and each time, that is recorded on the asset's timeline and in the audit log. Names come from your synced directory. Not in the directory means the person could not be matched. Concealed means the usage report did not give an identity that could be matched to your directory.

The weekly digest

Each week, after the first completed scan, the Command Center emails a digest to every administrator with an email address in your synced directory. Without directory sync, nobody receives it. See Directory in the portal.

The digest gives the posture score and its change from last week, new AI assets, new failing controls, findings by severity, how many assets are waiting for review, and how many times people were revealed. It is on by default. Turn it off with Send the weekly digest in Settings.

Evidence bundles

A bundle is a signed zip file of your posture: the inventory, the control results, the sources and the scan history.

On Evidence, select Request evidence. This needs the manager role.
Choose A single scan, or all the scans in the last 30, 90 or 365 days.
Select Request bundle. The row moves from Queued to Building to Ready.
Select Download. A bundle can be downloaded for 30 days, then reads Expired.

A bundle leaves people out unless you tick Include people. That option appears only with named visibility, for a member who holds the people detail role. Including people records a reveal on every asset's timeline and in the audit log, attributed to you. Such a bundle can be downloaded only by members who hold that role, and only while named visibility stays on.

A bundle that reads Failed says why: too large, no finished scan in that period, or a build failure you can retry.

Verifying a bundle

Each bundle contains a README.md with the full procedure, and it needs no AINOW code. An auditor:

Computes the SHA-256 of each file listed in manifest.json and compares it with the listed value.
Verifies manifest.sig, an Ed25519 signature over the hexadecimal SHA-256 of manifest.json.
Walks runs.jsonl, checking that each scan carries the previous scan's hash and that the last line matches the scan named in the manifest.
Compares the counts in the manifest with the files.

Check the signature against AINOW's published key, or one you obtained directly from AINOW. The key embedded in the bundle only proves the bundle is consistent with itself. README.md says where the key is published.

The Evidence page also mentions a verification script. You do not need it: the steps above are the full check. If your auditor does use it, a valid bundle reports OK WITH WARNINGS, because posture bundles are not anchored to the audit chain. Any failed check means the bundle should not be trusted.

Settings

Anyone with the viewer role can read Settings. Only administrators and managers can change it.

People visibility is Aggregate counts only (default) or Named people.
Send the weekly digest is on by default.
Per-person retention (days) defaults to 90 and can be set from 30 to 400. Older per-person usage is deleted, so the People tab cannot look back further.
Usage retention (days) defaults to 400 and can be set from 30 to 1,500. Older anonymous usage and daily rollups are deleted.
Inventory is AI assets only by default. With Every consented app, scans also list the consented apps that are not AI. To see them, set the inventory's filter to All apps.

Catalogue overrides is for managers. Hide a shipped entry stops a catalogue entry from matching for your organization, using the code shown as Catalogue entry on an asset. Add a local entry adds an app the catalogue does not know. Overrides apply from the next scan, and deleting one brings the shipped entry back.

For MSPs

Administrators of an MSP tenant see their customers' posture on the Customers page. A card headed AI posture across your customers shows how many customers are licensed, the Average posture score, Critical findings and Customers with a scored scan. The Posture column shows each customer's score, or Off when the customer is not licensed.

An administrator, or a member with the Pricing management role, can select License or Unlicense on a customer's row. See Partners and billing.