If you manage IT for a small business, you've probably heard some version of this from leadership: "Can you make sure employees aren't using unapproved AI tools?" And if you're honest, your answer is probably: you have no idea what AI tools your staff are already using.
That's the exact pain point driving discussions across IT and MSP forums right now, where admins are asking what actually works to restrict AI access for both on-prem and remote users. The honest answer is that blocking everything isn't realistic, and ignoring the problem isn't safe. You need a middle path: visibility first, policy second, enforcement last.
Why This Can't Wait
Two stories broke this month that should move AI governance up your priority list.
According to The Hacker News, malicious browser extensions can leverage Claude Dispatch to execute arbitrary code on an employee's machine. That means a tool your staff installed to "help with writing" or "summarize documents" could become a remote code execution vector — without the employee ever knowing.
At the same time, The Hacker News reported that researchers found a way to bypass ChatGPT's own Lockdown Mode and silently exfiltrate chat data without any user interaction. If you've been assuming the AI vendor's built-in security settings have you covered, this is proof that assumption is dangerous.
And zooming out, The Hacker News makes the broader case that the gap between AI capability and organizational readiness is accelerating faster than most IT teams can keep up with. Shadow AI isn't a future problem — it's already running on your network today.
Step 1: Find Out What's Actually Running
You can't govern what you can't see. Before writing a single policy, get a real inventory of AI usage across your network:
- Pull DNS and proxy logs for known AI domains (chat.openai.com, claude.ai, gemini.google.com, perplexity.ai, and the dozens of smaller tools employees discover on their own)
- Audit browser extensions across managed devices, since extensions are increasingly the delivery mechanism for AI-related exploits, as we covered in our browser extension security audit checklist
- Review OAuth app permissions in Microsoft 365 or Google Workspace for any AI tool that's been granted access to email, files, or calendars
- Ask department heads directly — shadow AI often starts as a sanctioned workaround to a slow approval process, similar to the patterns we outlined in our piece on shadow AI apps employees build on their own
Step 2: Tier Your AI Tools, Don't Blanket-Ban Them
A flat "no AI" policy usually fails because employees route around it, often through personal devices or personal accounts, which is worse for visibility than having them use an approved tool. Instead, create three tiers:
- Approved — vetted tools with business accounts, data retention controls, and admin oversight (think enterprise ChatGPT or Copilot with proper settings configured)
- Restricted — tools allowed for non-sensitive tasks only, with no company data permitted
- Blocked — tools with no enterprise controls, unclear data handling, or a history of security issues
This tiering approach is the same logic we recommend in our broader AI governance and shadow AI data leakage guide — it keeps productivity intact while closing the biggest gaps.
Step 3: Enforce at the Network Layer, Not Just Policy Documents
A written policy without technical enforcement is a suggestion. Use your firewall, DNS filtering, or secure web gateway to:
- Block known unapproved AI domains outright
- Flag new AI tool traffic for review before it becomes normalized
- Apply the same rules to remote and hybrid workers through endpoint-level DNS filtering, not just on-prem network rules
Step 4: Re-Audit on a Schedule
Shadow AI isn't a one-time cleanup. New tools launch weekly, and employees will keep finding workarounds. Set a recurring 30- or 60-day review of DNS logs, extension inventories, and OAuth grants so new risks don't sit unnoticed for months.
The Productivity Trap
The teams that struggle most with this are the ones that treat it as a security-versus-productivity tradeoff. It isn't. Giving staff one or two well-vetted, sanctioned AI tools with clear data rules almost always reduces shadow AI sprawl more effectively than a ban ever will. Employees want speed; give them an approved fast lane and they'll stop building their own.
Take Action
The exploits detailed above aren't hypothetical — they're active, documented, and targeting the exact kind of unmonitored AI usage happening on small business networks right now. Proactive scanning catches these exposures before an attacker does. Oscar Six Security's Radar gives you an affordable, fast way to check your network for unauthorized tools, risky extensions, and exposed access points for just $99 a scan. Check out our solutions and let's get ahead of it. Focus Forward. We've Got Your Six.