Finding the AI tools your staff already use

Staff adopt AI tools because they help: drafting emails, summarizing documents, cleaning up spreadsheets, transcribing meetings. The risk is not the productivity; it is what goes in. Client files pasted into a free chatbot, a browser extension that reads every page an employee opens, or a notetaker that records every client call are all disclosures to a third party your organization has no contract with.

The Canadian Centre for Cyber Security's generative AI guidance (ITSAP.00.041) lists privacy of data and loss of intellectual property among the risks of generative AI. The NIST AI Risk Management Framework organizes AI risk work into four functions: Govern, Map, Measure and Manage. Mapping, meaning knowing which AI systems are in use and for what, comes before you can sensibly govern them.

Start with your identity platform, because many AI tools let users sign in with their work Microsoft or Google account and request access to data while doing so.

1. 01

   Review enterprise applications

   In Microsoft Entra ID, the Enterprise apps list shows third-party applications that users or admins have consented to, and the permissions each holds. Look for AI assistants, notetakers, writing tools and 'productivity' apps with mail, calendar or file permissions.

2. 02

   Check sign-in logs

   Entra sign-in logs show which applications users are signing into with their work accounts and when. Filter by application to see how widely a tool is used.

3. 03

   Look at consent settings

   Microsoft documents user consent settings under Identity > Applications > Enterprise apps > Consent and permissions > User consent settings. Built-in options include not allowing user consent, or allowing it only for apps from verified publishers for permissions you classify as low impact.

4. 04

   Do the same in Google Workspace

   The Admin console's third-party app access controls list apps connected with Google accounts and the data they can reach.

AI browser extensions are easy to overlook and can be the most invasive: some can read and change everything on the pages a user visits, including web email, CRM records and online banking. If your devices are managed, endpoint management tools such as Microsoft Intune, or browser management policies for Edge and Chrome, can report installed extensions and restrict installation to an approved list.

If devices are not managed, the survey below and a quick look at a sample of machines will tell you what is common.

Many AI tools are simply websites, with no sign-in through your identity provider. Network and endpoint data catch these. Microsoft Defender for Cloud Apps, for example, offers cloud discovery that analyzes traffic logs against a catalog of over 31,000 cloud apps, ranked and scored on more than 90 risk factors, and can use data from Defender for Endpoint on managed devices. Firewalls and secure web gateways often have similar application reports.

Run discovery for a few weeks, then sort the results by number of users and by risk. You are looking for the handful of tools that most people use and the riskier ones that a few people use heavily.

Technical discovery misses personal devices and personal accounts. A short, explicitly no-blame survey fills the gap and tells you why people use these tools, which shapes a better policy.

• Which AI tools have you used for work in the past three months?
• Personal or work account?
• What kinds of tasks? (drafting, summarizing, research, code, meetings)
• Have you put client, staff or confidential information into any of them?
• What would you most want an approved tool to do?

If confidential client data has already gone into an unapproved tool, treat it as a potential privacy incident: find out what was shared, check the vendor's terms and deletion options, and assess with counsel whether there is a real risk of significant harm.

We run shadow AI discovery for organizations and help turn the results into an approved tool list and policy.

1. 01

   Build the inventory

   Tool, users, data involved, account type, vendor terms on retention and training.

2. 02

   Decide per tool

   Approve with a business account, approve with limits, replace with an approved alternative, or block.

3. 03

   Offer the better option

   Blocking without an approved alternative pushes use onto personal phones.

4. 04

   Tighten consent

   Restrict user consent to apps so new AI tools need approval.

5. 05

   Write the policy

   An AI acceptable use policy that names the approved tools and prohibited data.

6. 06

   Repeat quarterly

   New tools appear constantly; discovery is an ongoing check, not a project.