What AI tools are my employees using? How to find out

Obsidian Ridge offers the assessment described below. The inventory steps are our practical guidance; the offer description comes from our approved service wording.

Short answer: Ask employees which AI tools they use, then check the work devices and accounts because a conversation alone can miss extensions, desktop agents and personal accounts. Review installed apps, browser extensions, AI sign-ins on work accounts and the permissions each tool has. Record what each tool can open or change, including files, email and access to run commands.

Key facts

• How can you find out? Our recommendation, October 23, 2026: combine staff input with a device and account review; neither is a complete inventory on its own. The published assessment scope includes device AI tools and access.
• Tool coverage: The service describes browser, desktop and command-line AI tools on staff devices. Assessment scope, checked October 23, 2026.
• Access coverage: The service examines the client data and systems those tools can reach. Assessment scope, checked October 23, 2026.
• Findings: The service flags risky actions and includes a CISSP review of findings and recommended fixes. Assessment scope, checked October 23, 2026.
• Terms: The approved description below includes the engagement length, eligibility and separate implementation boundary. Assessment terms, checked October 23, 2026.

What should we ask employees first?

Ask about the task before asking for a product name. Our suggested opening is: “Which tools help you write, summarize, search, build spreadsheets or automate work?” Invite staff to include occasional use and tools they tried but no longer need.

Make it clear that you are collecting an inventory. Ask which account they use, whether they installed anything, and what information they give the tool. Use a blank example document if a demonstration would otherwise expose client information.

A useful staff response is a lead for the technical review. Do not mark a device as checked merely because its owner remembered a familiar chatbot. Likewise, do not treat a staff member's uncertainty as evidence of wrongdoing. Record the uncertainty and inspect the relevant work setup with the person who administers it.

What should the device and account checks include?

Inspect the places where tools are installed, connected and authorized, within the access the business has agreed to review. Our recommended inventory has a row for each tool and account combination, because a name alone does not explain the access.

Have the administrator show the evidence behind a row. Record “not checked” when a device, account or log is unavailable. Do not collect passwords or ask staff to expose private personal conversations to fill a spreadsheet.

Our command-safety guide and fake-download analysis explain why a tool's name is not enough to decide whether its installation route was safe. The inventory should record where an unfamiliar app came from and who approved it, without trying to reproduce a suspicious installation.

How do we find out what an AI tool can reach?

Review its permissions and connections, then verify the allowed actions with safe test material. Separate reading a file from changing it, drafting an email from sending it, and suggesting a command from executing it.

Our recommendation is to use a simple access record: source of the information, account used, allowed action, approving person and test result. If a tool is connected to a whole drive but the approved task needs only a folder of public material, ask the administrator to review that mismatch.

Keep the observed result beside the intended rule. An approval document that says “public information only” is not proof that the connected account cannot open client files. Conversely, seeing a permission does not prove that data was actually taken. Distinguish potential access, observed activity and unanswered questions in the report.

What should the owner do with the inventory?

Assign an owner and decision to each finding. Our recommended outcomes are approve, restrict, remove through the normal change process, or investigate further. Give unresolved items a named person and a next review date.

Do not remove a tool blindly if staff rely on it for an active business process. Understand the dependency, agree the change and provide a permitted replacement route when needed. The inventory should help the practice make decisions it can carry out.

Tell staff which tools and accounts are approved, and how to request a new one. Repeat the review when the business adds a tool, changes access or finds a gap. The goal is a maintained record of decisions, not a spreadsheet that becomes an assurance the business cannot support.

Where does Obsidian Ridge's assessment fit?

It provides the AI-tool, access and risky-action review described in the approved offer below. Agree which devices and accounts are in scope before work starts, and ask how gaps in visibility will be reported.

Use the results to decide what to address first. Do not read a finding-free sample as proof that no employee has ever used an unapproved tool. Discovery informs the next decision; implementation and paid tools remain separate under the offer's own terms.

Sources

• Obsidian Ridge AI security assessment: service scope and approved first FAQ, checked October 23, 2026.
• The inventory table and review sequence are Obsidian Ridge's recommended working process, not extra promised assessment deliverables.

Related reading

• Is it safe to run a Terminal command ChatGPT gives you?
• How to stop employees pasting client data into ChatGPT
• Fake AI app downloads: how to spot a poisoned Claude or ChatGPT ad
• AI agents on work laptops: what can they reach?
• Compare AI security assessments
• Write an AI usage policy