Cybersecurity
Oct 26, 2026
Oct 30, 2026
•
12 min read
With CloudConnexa's Cyber Shield, you add the domains of unapproved AI services to a Domain Block List in a Shield policy, then assign that policy to the user groups you choose. Approved AI tools stay available, unapproved ones are blocked, and the policy follows users wherever they work.
This guide explains what shadow AI is, why an AI policy alone doesn't stop it, and how to enforce your policy step by step, including for AI agents running on servers. It's part of our agentic AI security guide for SMB and mid-market IT teams.
Key takeaways
- Shadow AI is any AI tool or service used for work without approval from IT or security.
- It's common and costly: in IBM's 2025 Cost of a Data Breach research, one in five organizations reported a breach due to shadow AI, and high levels of shadow AI added an average of $670,000 to breach costs.
- A written AI acceptable use policy is the starting point. A network-level control is what enforces it automatically.
- In CloudConnexa, a Cyber Shield Domain Block List works as a custom category for unapproved AI services. It blocks each listed domain and its subdomains.
- AI agents need the same control. Agents on servers usually connect as Hosts, which always use the Default Policy, so the block list belongs there too.
What is shadow AI?
Shadow AI is the use of AI tools and services that your IT or security team hasn't approved. It's usually well-intentioned: an employee pastes a customer email into a free AI chatbot to draft a reply, uploads a spreadsheet to an AI tool to summarize it, or installs an AI browser extension to save time.
The problem isn't AI itself. It's that company data ends up in services you haven't vetted, under terms you haven't agreed to, with no record of what was shared.
Shadow AI vs. shadow IT
Shadow AI is a subset of shadow IT, the broader use of any unapproved software or service. It gets its own name because of how AI tools are used: people paste in or upload the data itself, and the service may keep or learn from it under its own terms. More and more, it isn't only people doing it. AI agents can call external AI services on their own.
|
Shadow IT |
Shadow AI |
|
|---|---|---|
|
What it is |
Any unapproved app, cloud service, or device |
Unapproved AI chatbots, writing tools, file processors, extensions, and AI APIs |
|
Typical risk |
Unmanaged accounts and data stored in unknown places |
Sensitive data pasted or uploaded directly into a third-party model |
|
Who does it |
People |
People and AI agents |
Learn more about shadow AI and shadow IT in our recent 2026 OpenVPN [Con]nected panel here.
Shadow AI risks for SMBs and mid-market companies
- Data leaves your control. Customer data, source code, contracts, and financials can be sent to services that may retain or use it under their own terms. This is the most common form of AI data leakage.
- Compliance exposure. If you handle regulated data, sending it to an unapproved service can create compliance problems and gaps in your audit trail.
- No visibility. You can't protect or audit data flows you don't know about.
- AI agents can do it too. An agent that's been manipulated, or simply built carelessly, can send data to an external AI service just as easily as a person can.
Smaller teams are often more exposed than large enterprises because they adopt new tools quickly and have fewer controls in place. IBM's research found that 63% of breached organizations had no AI governance policy or were still developing one.
Why an AI acceptable use policy alone isn't enough
Most organizations start with an AI acceptable use policy that lists approved tools. That's an important first step, and frameworks like the NIST AI Risk Management Framework start with governance for good reason. But a policy depends on everyone remembering and following it, every time. A network-level control turns the policy into something that's enforced automatically.
The goal isn't to block AI. It's to make the approved path easy and the unapproved path unavailable.
How CloudConnexa controls shadow AI
Cyber Shield is CloudConnexa's built-in content filtering and IDS/IPS. Its Domain Filtering works at the DNS layer and sorts domains into 43 content categories. There's no built-in AI category, so you use a Domain Block List instead. According to the CloudConnexa docs, the block list functions as a custom domain filtering category. That's an advantage: the list matches your own AI policy exactly, not a vendor's idea of what counts as AI.
Three things to know before you start:
- Block lists live inside Shield policies. Each Shield policy has its own Domain Filtering settings, Domain Block List, and Domain Allow List.
- Each user group belongs to exactly one policy. You assign user groups to a policy. Groups you don't assign use the Default Policy.
- Networks and Hosts always use the Default Policy. You can't assign them to a custom policy. This matters for AI agents running on servers.
Because filtering happens at the DNS layer, the CloudConnexa docs note that protection applies even when a user's internet traffic doesn't pass through your WPC. Your AI policy follows people whether they work from the office, home, or anywhere else.
How to block unapproved AI tools: step by step
Step 1: Decide what's approved
Agree on which AI services your organization approves, for example a business AI subscription with data protection terms. Write this down as part of your AI acceptable use policy, along with who can approve new tools.
Step 2: List the unapproved AI services
List the domains of common AI services that aren't approved, including chatbots, AI writing tools, and AI file-processing services. Include the domains each service uses for its app and its API where relevant. A block list entry covers the domain and all its subdomains, so you usually need only the root domain. Keep the master list in a .txt file, one domain per line. You'll reuse it in every policy.
Step 3: Add the list to a Shield policy
In the CloudConnexa Administration portal, go to Shield → Policies and create or edit a policy with Domain Filtering turned on. Name it something clear, such as Unapproved GenAI. Then:
- Select the Domain Block List tab and turn on Enable Block List.
- Enter the domains from Step 2, one per line, or select Upload List to load your .txt file.
- Check that none of your approved AI services are on the list.
- Select Save changes.
Each list holds up to 150,000 characters. Upload List replaces the current list rather than adding to it, which is why keeping one master .txt file makes updates simple.
If an approved service shares a parent domain with a blocked one, you can exempt the specific subdomain with the Domain Allow List, since the more specific entry takes precedence.
Step 4: Assign the policy to employee groups
From the policy's Overview tab, select Assign user groups and choose your employee user groups. Because each group can belong to only one policy, add the same block list to any other custom policy your employee groups already use.
If some teams have an approved exception, for example a team evaluating a new tool, keep their group in a separate policy without the block list. The number of custom Shield policies depends on your plan (1 on Starter, 5 on Essential, 10 on Premium, and 100 on Enterprise; the Default Policy doesn't count).
Step 5: Apply the same block to AI agents
How you do this depends on how the agent connects to CloudConnexa:
- Agents on servers or VMs connected with a Host Connector: Hosts always use the Default Policy. Add your unapproved AI list to the Default Policy's Domain Block List.
- Agents connected as users or devices in a user group: assign that group to a policy with the block list, just like employee groups.
Either way, an agent connected through CloudConnexa can't reach unapproved AI services, which limits where a manipulated agent can send company data. For the rest of an agent's access, see how to give AI agents least-privilege access.
Step 6: Test, monitor, and communicate
Confirm that approved AI tools work and unapproved ones are blocked. If a domain still loads right after you save, local DNS caching may be the reason, so test again after the cache clears.
Then use the Cyber Shield Top 10 dashboard to see which users, networks, and hosts are behind blocked events. To keep a record for audits, stream Cyber Shield log events to your SIEM, including domain-blocked events.
Finally, tell employees what's approved and how to request a new tool. People are much more likely to follow a policy when the approved option is easy to use.
How to detect shadow AI with blocked-event data
Blocked events are more than a security log. Each one is a signal that someone needs an AI tool your organization hasn't provided yet. Review them regularly to:
- Spot demand. Repeated blocks for the same type of tool suggest a team needs an approved alternative.
- Find agents that misbehave. A Host or agent group that keeps hitting blocked AI domains may be misconfigured or manipulated.
- Prove enforcement. Domain-blocked events in your SIEM show auditors that your AI policy is actually enforced.
Keeping your list up to date
New AI services launch constantly, so a block list needs regular maintenance:
- Review the list on a regular schedule, such as monthly.
- Add new services when employees request them, when they show up in blocked-event reviews, or when you become aware of them.
- Update your master .txt file first, then upload it to every policy that uses it. To automate this, use the CloudConnexa API's Cyber Shield domain filtering endpoints.
- Give employees a simple way to request approval for a new tool, so blocking doesn't push them toward workarounds.
What this approach does and doesn't cover
Cyber Shield controls which AI services users and agents can reach, by domain. It doesn't inspect the content of prompts or what's shared with approved AI services. Control what can be shared with approved services through those services' own business settings and your data handling policy.
Shadow AI control checklist
- Approved AI services are documented in an AI acceptable use policy
- Unapproved AI domains are kept in one master list
- The list is in the Domain Block List of every Shield policy that covers employees
- The list is in the Default Policy, so Hosts and Networks (including AI agents on servers) are covered
- Exception teams sit in a separate policy
- Approved services are tested and confirmed working
- Blocked events are reviewed and streamed to your SIEM
- The list is reviewed on a regular schedule
- Employees know how to request a new tool
Frequently asked questions
What is shadow AI?
Shadow AI is the use of AI tools and services that an organization's IT or security team hasn't approved. It creates risk because company data can be sent to services the organization hasn't vetted.
What's the difference between shadow AI and shadow IT?
Shadow IT is any unapproved software or service. Shadow AI is the subset involving AI tools, where people or agents send data directly to a third-party model that may keep or use it under its own terms.
Is using a free AI chatbot at work shadow AI?
Yes, if your organization hasn't approved it. The same chatbot can be approved when used through a business subscription with data protection terms that your organization has agreed to.
How do I block unapproved AI tools at work?
Create a list of unapproved AI service domains and block them at the network level. In CloudConnexa, add the domains to the Domain Block List of a Cyber Shield policy and assign that policy to the user groups you choose, while leaving approved AI services accessible.
How do I detect shadow AI?
Review blocked-event data. In CloudConnexa, the Cyber Shield Top 10 dashboard shows which users, networks, and hosts are behind blocked events, and Log Streaming can send domain-blocked events to your SIEM.
Does CloudConnexa have a built-in AI category?
No. CloudConnexa doesn't have a built-in AI category. You use a Domain Block List, which functions as a custom category, so the list matches your own AI policy.
Can I block AI services for AI agents, not just employees?
Yes. Agents that connect as Hosts use the Default Policy, so add the block list there. Agents that connect as users or devices in a user group get that group's Shield policy.
Does blocking shadow AI stop data leaks to approved AI tools?
No. Blocking controls which AI services can be reached. It doesn't inspect what's shared with approved services. Use approved services' business data protection settings and your data handling policy to control that.
.png)