AegisAI Perspectives · 125
Find the AI tools employees use from the sign-up, verification or billing emails in Microsoft 365 or Google Workspace, with no agent or proxy to install.
See it on your own Google Workspace or Microsoft 365 tenant. No MX change.
Or get the Q3 2026 State of the AI Attack report (free, instant download)
Not ready for a demo? Check your domain (free, about 10 seconds).
Search work mail for sign-up, verification or billing emails from AI tools: each names the tool and who adopted it, with no agent or proxy. In 2024, 78% of AI users brought their own AI tools to work12.
AegisAI reads the sign-up mail it already sees and builds an inventory of every app adopted with a company identity, AI tools first1. If you run IT for a company of 20 to 500 people on Microsoft 365 or Google Workspace, with a lean IT or security team and no SOC, you can run the same search by hand from the admin consoles you already have. The product is shadow IT and AI discovery; the buying criteria are in how to choose shadow IT discovery tools.
Because an AI tool confirms a new account by email. The sign-up, verification, team invite or receipt lands in the mailbox of the person who used a work address, so the inbox records the tool and the person, with a date. Email catches the sign-up even when the app never touched single sign-on (SSO) or your identity provider8.
That matters most for free tiers. Proxies, cloud access security brokers (CASBs) and single sign-on (SSO) reports miss the free-tier sign-up made with a work address9. Those free sign-ups carry cost: organizations with high levels of shadow AI observed an average of $670,000 in higher breach costs, and one in five organizations reported a breach due to shadow AI14.
Treat every AI sign-up email as a row in an inventory, not as noise.
Short on time? Watch the explainer, or book a demo.
The search mechanics for any SaaS app, on both platforms, are in how to discover SaaS apps employees signed up for with their work email. These steps narrow that search to AI tools and turn the result into a decision.
Write the subject words AI tools use at each stage: sign-up and verify for a new account, invited you for a team workspace, receipt or subscription for a paid seat.
In Microsoft 365, run message trace over the longest window: up to 90 days of history, 10 days at a time17.
Export the Summary report: for each message it lists the date, sender, recipient, subject and status.
Group the rows by sender domain and keep only the domains that belong to AI tools. One domain becomes one row.
Mark each row free or paid: a receipt means a paid seat, a sign-up with no receipt means a free tier.
Check consent for each AI tool. In Microsoft Entra, user consent is initiated when a user signs in to an application18, and admins can view the permissions granted to an app20.
In Google Workspace, read each AI app's access level for the organization or organizational unit: Trusted, Limited, Specific Google data or Blocked22.
Add the owner, the first date seen and the data the tool could receive to every row.
Run steps 1 to 5 this week; they need no new software.
Two things. First, accounts opened with a personal address: sign-up mail shows adoption with a company identity4, so a personal account never reaches your tenant. Second, content: the search finds which tools people adopted and when, not what they type into them5.
Consent records cover AI tools that connect to the tenant, because a user signing in with a company account starts a consent. Network logs are the other route, and they need the infrastructure this method avoids. Microsoft's cloud discovery analyzes traffic logs against a catalog of over 31,000 cloud apps23, but its snapshot reports use traffic logs you manually upload from your firewalls and proxies.
Cover personal accounts in your AI policy, and use consent records as the second signal.
Sort each tool into one of three groups. Approved means the company buys a business plan and moves users onto it. Restricted means the tool stays for public or non-confidential material only. Prohibited means access is removed. Ask each user what the tool does for them before you decide: the answer often names the approved tool you should buy.
Write the rule down. Of breached organizations, 63% either don't have an AI governance policy or are still developing one15.
Keep the sheet as the record: tool, owner, first date seen, status, plus the date of each decision.
Act in the admin consoles. Revocation happens in your identity provider or admin console7. In Microsoft Entra, a Privileged Administrator can configure whether nonadministrator users are allowed to grant user consent to an application19. Revoking the current granted permission doesn't stop users from re-consenting21, so change the consent setting too. In Google Workspace, set a prohibited app to Blocked.
Send each user a two-line note naming the tool and its new status before you block anything.
Look for four things: nothing installed on devices or in the network path, coverage of free sign-ups made with a work address, a separate list of AI tools, plus who adopted each one and when. The full criteria are in how to choose shadow IT discovery tools. Hold every product you trial to the same four tests.
| Criterion | Mailbox sign-up search | Consent records (Entra, Workspace) | Network traffic logs | AegisAI |
|---|---|---|---|---|
| Needs an agent or proxy | No | No | Logs come from firewalls and proxies | No new agent or proxy |
| Sees a free sign-up with a work address and no SSO | Yes | Only when the user signs in with a company account | Only traffic in the uploaded logs | Yes, from sign-up mail |
| Sees accounts opened with a personal address | No | No | Only traffic in the uploaded logs | No edge. It sees adoption with a company identity |
| Shows what people type into the tool | No | No | No | No edge. It finds which tools people adopted and when |
| Removes access | No | Admins can review and revoke permissions | No | No edge. Revocation happens in your identity provider or admin console |
In the last three rows the product column has no edge: it does not see personal accounts, prompts or uploaded files, and it does not revoke access itself. Fill this table for your own tenant before you pick a method.
AegisAI reads the sign-up mail it already sees and builds an inventory of every app adopted with a company identity, AI tools first. Email catches the sign-up even when the app never touched single sign-on (SSO) or your identity provider. It finds which tools people adopted and when, not what they type into them. Revocation happens in your identity provider or admin console.
Illustrative example, figures are fictional: Larchmont Labs (fictional) has 140 people on Google Workspace and one IT lead. The search over the last 30 days returns 23 messages from 9 sender domains. Six domains belong to AI tools: a meeting note taker, two writing assistants, a code assistant, an image generator, plus a chatbot. Two of the six sent receipts, so they are paid seats. The access check shows the note taker connected to company accounts for 11 users. IT approves the code assistant on a business plan, restricts the writing assistants to public material, asks the 11 users what they record, then sets the note taker to Blocked. The next month's search finds one new AI sign-up.
Run this once on your own tenant and count the AI rows.
Track three numbers, each from a source you can export.
Put the three figures in one row on day 30 and set the next search date.
Search your mail logs for AI sign-up, verification or billing emails, group them by sender domain, then check consent records. AegisAI builds that inventory from mail it already reads, AI tools first. To see that list for your own tenant, ask for a walkthrough on your own mail.
The best shadow AI discovery tools meet four criteria: nothing to install on devices or the network, coverage of free sign-ups made with a work address, AI tools listed apart from other apps, and the person and date for each adoption. The full checklist is in how to choose shadow IT discovery tools.
No. Sign-up mail shows adoption with a company identity. Cover personal accounts in your AI policy and ask about them when you review the list.
In a 2024 survey, 78% of AI users brought their own AI tools to work. One in five organizations reported a breach due to shadow AI.
Not on its own. Revoking the current granted permission doesn't stop users from re-consenting. Change who may grant user consent, or set the app to Blocked in Google Workspace.
