Solutions / Shadow AI

See every AI tool. Control every AI action.

Endpoint Guard runs on Windows and Mac laptops. It finds the AI tools your people use, shows what they use them for, and checks every action an AI assistant takes on the laptop.

Colleagues discussing their work around laptops
Why Endpoint Guard

Four things that matter.

01

Every tool call on the laptop

Shell commands, cloud CLI calls, scripts, file writes and MCP tools. Checked before they run.

02

Policy in plain English

Write the rule the way you would say it: "Never delete anything in production AWS."

03

Every AI tool, and what it is for

Apps, coding assistants and local models. Work use and personal use.

04

Deployed by your MDM

Jamf, Rippling or Microsoft Intune. Windows and macOS.

Tool call governance

Govern every AI action, not only MCP.

Most AI security tools check MCP tool calls. AI assistants also run shell commands, call cloud CLIs and write scripts, using the employee's own credentials. Endpoint Guard checks all of them.

What gets checkedMCP-only toolsCortega
MCP tool calls✓✓
Shell commands–✓
Cloud CLI calls, such as AWS–✓
Scripts and SDK code the assistant writes–✓
Access to credential files–✓
Your own rules, in plain English–✓

You write the policy

"Never change anything in our production AWS account. Reading is fine."

"Never download a script and run it."

"Keep AI assistants away from cloud credential files."

"Ask the developer before a staging deploy."

Cortega applies it

aws ec2 terminate-instances --profile prodBlocked
aws s3 ls --profile prodAllowed
curl https://… | shBlocked
cat ~/.aws/credentialsBlocked
aws lambda update-function-code --profile stagingAsks first

A blocked command does not run. The assistant is told why and does not try another way. An approver can allow it for an hour. Works with Claude Code, Codex and Cursor.

Visibility

See what AI your people use, and what for.

01

Every AI app

ChatGPT, Claude, Copilot, Gemini, Cursor and more. Local models such as Ollama, too.

02

Work or personal

See which AI use is for work and which is not.

03

Usage and cost

Tokens, cost and lines of code for each person, even on their own subscriptions.

04

Prompts and code

Read prompts when your policy allows. List the code repositories on each laptop. The scan never uploads code.

Data protection

Stop data leaks on the laptop.

01

Block an app

Choose watch, protect or block for each AI app.

02

Mask sensitive data

Personal data, health data and keys are masked or blocked before they leave.

03

Works offline

Checks run on the laptop. They keep working when Cortega cannot be reached.

In the console

One place to set the rules.

App policy

Set policy for each supported AI application.

Choose how Endpoint Guard treats traffic from an app, independently of model routing.

Observe
Inspect and report configured traffic.
Police
Apply configured local controls.
Block
Stop the app’s governed connections.
Privacy
The privacy setting can limit what message content reaches Cortega.
Set policy for each supported AI application. in the Cortega console
Actual product screenshot · Current application policies; cropped to the policy controls. Select to enlarge.
Deployment

Deploy with the MDM you already use.

  • Jamf
  • Rippling
  • Microsoft Intune
  • Windows
  • macOS
How it works

Endpoint Guard architecture

An agent on each laptop applies your app policies and checks. Results go to the Cortega console.

Endpoint Guard applies app policy on macOS. Block stops an app; observe inspects and reports; police applies local guardrails before forwarding allowed traffic to the original destination. Other domains remain encrypted. Responses return through configured local checks. Supported assistants have a separate optional route through AIBG. The console receives findings and sends configuration; AI Verifier and Cortega Agents are separate integrations.
Teal: outbound traffic. Gray: responses. Dashed: reporting and configuration. App inspection covers supported, configured domains; AIBG controls apply to traffic routed through AIBG.
Inspection, recording, and optional analysis

Observe and police modes inspect configured app domains. Police mode adds local guardrails; each guardrail can log, mask, or reject content on requests, responses, or both. Domains outside inspection scope remain encrypted.

Set recording and data-handling policies for endpoint observability. AI Verifier and Cortega Agents are enabled separately for selected traffic; they are not part of the local guardrail execution path.

Shadow AI

Find the AI tools on your laptops.

Start with a few laptops. Add your MDM when you are ready.