Table of contents
- TLDR
- What is shadow AI?
- Which AI tools are actually showing up?
- Why shadow AI matters for your clients
- How to detect shadow AI on endpoints with a PowerShell script
- What a script can’t catch
- Script vs. paid AI usage detection: when to pay for more
- What to do once you find AI usage
- Your client might actually want their team using AI
- How Syncro fits
- Conclusion
- Frequently Asked Questions About Shadow AI
TLDR
- Shadow AI is the unauthorized or unknown use of AI tools on a business’s computers. For most SMB clients, that means free ChatGPT in a browser and Copilot.
- The biggest risk is an employee handing sensitive data to an AI tool, especially in regulated environments governed by HIPAA, PCI or GDPR.
- A PowerShell script can check processes, browser history and DNS traffic against a list of known AI tools and tell you which tools were used, by whom and on which device.
- A script answers “was it accessed?” Paid security tools answer “what was done with it?” and can block it. Start with the script, then decide what to pay for.
- Use what you find to start a conversation: at a QBR, as a lead-in to security services, or as a paid service or value-add.
What is shadow AI?
Your clients’ employees are already using AI at work. The question is whether anyone at the business knows about it.
Shadow AI is the unauthorized or unknown use of AI tools on a business’s computers. It’s the AI version of shadow IT: tools nobody approved, running on devices you’re responsible for.
I started paying closer attention to this when I noticed how many security vendors were suddenly marketing “AI usage detection” as a feature. Vendors don’t all build for the same problem at once by accident. They’re responding to demand from MSPs and their clients.
Which AI tools are actually showing up?
For the most part, it’s the usual suspects. On SMB endpoints, the tools showing up most are the free or included ones:
- ChatGPT, almost always through a web browser rather than a desktop app
- Microsoft Copilot, because it’s already sitting in Windows, Edge and Microsoft 365
- Other chat assistants like Claude and Gemini, plus AI browser extensions that ride along with everything a user types
That matters for detection. If you only look at installed software, you’ll miss most of it, because a browser tab doesn’t show up in an app inventory.
Why shadow AI matters for your clients
The most common risk is straightforward: someone gives an AI tool access to sensitive data.
Picture a front desk employee at a medical practice pasting visit notes into a free chatbot to tidy up a referral letter. They’re trying to save time. They’ve also just sent protected health information to a third party with no business associate agreement in place. For clients under HIPAA, that’s a compliance problem, not just a security one.
Shadow AI risk is a data handling risk first: the danger is what employees paste in, not the tool itself. The same goes for client contracts, financial records, credentials and anything covered by a client’s own compliance requirements.
How to detect shadow AI on endpoints with a PowerShell script
I wrote a PowerShell script to check for AI usage, and I’ve added it to the Syncro Community Script Library so you can import it and run it yourself.
What the script checks
The script uses an embedded table of common AI tool names and checks three places for matches:
- Running processes: is an AI app or assistant running right now?
- Browser history: has a user visited an AI tool’s website, and in which browser?
- DNS traffic: is the machine resolving domains that belong to AI services?
What the output looks like
Example: a run on one workstation might return something like this:
ChatGPT (OpenAI)
- Website visited by USERNAME (Edge)
Microsoft Copilot
- Running right nowThat’s enough to tell you which tool, which user and how it was accessed.
Pro Tip: the AI tool table is meant to be edited. Add your own keyword matches for tools your clients use, industry-specific AI apps or anything new that pops up. The list gets more useful the more you personalize it.
What a script can’t catch
A script is a great first look, but be honest with yourself and your clients about its limits:
- Personal devices. If an employee uses ChatGPT on their personal phone, nothing on the work endpoint will see it.
- Shadow IT assets. Unmanaged laptops or devices you don’t have an agent on are invisible to any endpoint script.
- Masking techniques. Users who deliberately hide their activity with more advanced methods can get past basic checks.
- History. A script is limited in how far back it can look. Paid tools are better at tracking and finding older access.
Script vs. paid AI usage detection: when to pay for more
At the base level, a script and the vendor features aren’t that different. They both look for known AI tools on the endpoint. The difference is depth.
Paid security tools can catch more advanced techniques that try to hide access. More importantly, they go beyond access. They have the potential to see what’s being done with AI tools in real time and block risky activity before it happens. A tool like Bitdefender might flag that an MCP server was accessed through Command Prompt, or recognize that PII or credentials were provided to an AI tool.
| Capability | PowerShell script | Paid security tools (ThreatDown, Bitdefender, etc.) | Microsoft 365 Shadow AI page (preview) |
|---|---|---|---|
| Finds known AI tools on managed endpoints | Yes | Yes | Yes, for a set list of agents and desktop apps |
| Browser-based AI use | Yes, via browser history and DNS | Yes | Limited to supported agents |
| Shows what data was shared | No | Some tools flag PII or credentials | No |
| Historical lookback | Limited | Stronger | Shows first and most recent activity |
| Blocks AI tools | No | Yes, in many tools | Yes, for some agents via Intune |
| Cost and requirements | Free, runs through your RMM | Paid add-on or tier | Microsoft 365 E5, Defender for Endpoint, Intune and Global Secure Access |
The script is enough when you need a quick answer to “is anyone using AI here?” Use it for a baseline, a prospect assessment or a client that hasn’t asked for more.
Pay for more when the client is regulated, when they need proof of what data left the building, or when they want AI blocked rather than just reported.
What to do once you find AI usage
Finding AI usage is a conversation starter.
Bring it to the QBR
Put the results in front of the client at their next QBR. “Here are the AI tools your team used this quarter” lands a lot harder than a generic slide about AI risk. It also works for prospecting: run it during an assessment and show a prospect what’s happening on their network today.
Move from visibility to control
If you find unauthorized AI usage, it’s a natural lead-in to security services like firewalls, DNS filtering, MDR and XDR. Those are the controls that actually let a client restrict or block access, instead of just knowing about it. If the client wants a tighter approach, application whitelisting is another option for controlling what runs on their endpoints.
Package it as a service or a value-add
Recurring AI usage reporting can be sold on its own or bundled into an existing package as a value-add. Run the check on a schedule, review the results monthly and report on it. It’s the same play MSPs already run with identity security as a managed service.
Put a policy in place before you monitor
Before you start scanning, make sure the client has an acceptable use policy that covers AI and that employees know company devices are monitored. Monitoring rules vary by state and country, so have the client loop in their own HR or legal team on what to disclose.
Your client might actually want their team using AI
It might seem counterintuitive, but not every AI tool you find is shadow AI or something to shut down. A lot of small businesses want their employees using AI. What they want is for it to happen with the right tools and the right data. The same detection script helps there too. It shows which tools people are already reaching for, so the client can approve and pay for the ones that work, and steer people away from the risky ones.
That’s how you go from the MSP who says “no” to the trusted technology advisor who helps clients adopt AI safely.
How Syncro fits
Syncro’s RMM scripting lets you run a detection script like this across every client from one place. You can run a script on a single asset, a saved search, a whole customer or a policy folder, and schedule it anywhere from once to monthly, with daily as the recommended ceiling for scheduled runs. Results can show up on each asset’s Scripts tab or be written to the activity log or custom asset fields, so you can spot what changed between QBRs.
Start by importing the AI usage script from the Community Script Library, test it on a few machines, then add it to a policy so it runs on its own. See how Syncro’s scripting and automation works.
Conclusion
Most of the shadow AI on SMB endpoints is free ChatGPT and Copilot, and the tool itself usually isn’t the problem. The sensitive data that ends up in it is, especially for clients governed by HIPAA, PCI or GDPR.
A PowerShell script gets you a baseline today: which tools, which users, which devices. Paid security tools go deeper when a client needs more detail or wants access blocked. Either way, what you find is worth bringing to the client, whether that means tightening controls or helping them adopt AI on purpose.
If you only do one thing about shadow AI this month, get familiar with the common AI tool names and how people reach them. Everything else, from detection to policy to the QBR conversation, builds on that.
Ready to run AI usage checks across every client from one console? See Syncro pricing and start a free trial.
Frequently Asked Questions About Shadow AI
Shadow AI is the unauthorized or unknown use of AI tools on a business’s computers. For SMBs, it’s usually free tools like ChatGPT in a web browser and Microsoft Copilot that employees use without IT’s approval or knowledge.
MSPs are responsible for the endpoints and data where shadow AI happens. When an employee pastes sensitive data into an unapproved AI tool, the client faces data leak and compliance risk, and the MSP is the one expected to know about it.
The biggest risk is employees sharing sensitive data with AI tools, such as patient information, client records, financial data or credentials. For businesses in regulated environments, that can create a compliance problem on top of a security one.
You can detect shadow AI by checking running processes, browser history and DNS traffic for known AI tool names and domains. A PowerShell script run through an RMM can do this across many devices, and paid security tools add deeper visibility and blocking.
Yes. A PowerShell script can detect ChatGPT usage by checking browser history for visits to OpenAI’s site and DNS traffic for its domains. It reports whether ChatGPT was accessed and by which user, but not what was typed into it.
A script can’t see AI use on personal phones, unmanaged devices or shadow IT assets, and it can miss users who deliberately mask their activity. It also has limited historical lookback compared with paid security tools.
It depends on the client. Many small businesses want their employees using AI, so the goal is usually to approve the right tools and restrict risky ones rather than block everything. Detection shows which tools people already use, which makes that decision easier.
MSPs can run AI usage checks on a schedule, report the results at QBRs and use findings to recommend controls like DNS filtering, firewalls, MDR and XDR. Recurring AI usage reporting can also be sold as its own managed service or bundled into an existing package as a value-add.
Share
Related Resources
