---
title: How do you find AI steps hidden inside Zapier, Make and Power Automate workflows?
url: https://opencontroller.lyzr.com/blog/find-ai-steps-in-zapier-make-and-power-automate
published: 2026-11-14
updated: 2026-11-14
author: Lyzr
---

# How do you find AI steps hidden inside Zapier, Make and Power Automate workflows?

**Pull the workflow inventory from each tool's admin API, then search every workflow for AI modules and model-provider connections. In Power Automate that means the Power Platform inventory's Connectors column; in Make it means each scenario's usedPackages and blueprint; in Zapier it means the Apps page and the connections behind each Zap. Then match the API keys in those connections back to the keys you already know about.**

No-code workflows are one of the quietest places AI enters a company. A finance analyst adds a "summarise this email" step to a Zap. A sales ops lead drops an OpenAI module into a Make scenario. Nobody files a ticket, because it does not feel like building an AI agent. But each of those steps sends business data to a model, often with a personal API key. This guide is part of our [shadow AI agent discovery guide](/blog/how-to-find-shadow-ai-agents).

## Where do AI steps hide in each tool?

AI shows up in three forms in automation platforms: a built-in AI action sold by the platform, a module or connector for a model provider, and a raw HTTP call to a model API.

| Tool | Built-in AI steps | Model-provider steps | Where to look as an admin |
|---|---|---|---|
| Power Automate | AI Builder actions, including "Run a prompt"; Copilot Studio agent flows | Azure OpenAI and other AI connectors; HTTP action to any model endpoint | Power Platform inventory (Connectors column and tab), Inventory API |
| Make | Make's own AI agent and AI toolkit modules | OpenAI, Anthropic Claude and other AI apps; HTTP module | Scenarios API (`usedPackages`, `usedModules`, blueprint); audit logs on Enterprise |
| Zapier | AI by Zapier steps; Zapier Agents | ChatGPT (OpenAI), Anthropic and other AI apps; Webhooks or API steps | Enterprise Apps page (connections and affected Zaps); app allow list |

The third form, a generic HTTP step that posts to a model endpoint, is the one most scans miss. Search step configurations for model host names such as `api.openai.com`, `api.anthropic.com`, `*.openai.azure.com` and `generativelanguage.googleapis.com`.

## How do you scan Power Automate?

Microsoft's [Power Platform inventory](https://learn.microsoft.com/en-us/power-platform/admin/power-platform-inventory) covers agents, apps and flows, including cloud flows and Copilot Studio agent flows. For each flow it records the connectors and operations used, plus the trigger connector.

1. Open Power Platform inventory in the admin center, or query it with the "Query Power Platform resources" action, the Power Platform CLI (`pac resource-query query-resources`) or Azure Resource Graph.
2. Filter flows whose connectors include AI Builder, Azure OpenAI or any AI-category connector.
3. Separately filter flows that use the HTTP connector, then inspect those definitions for model host names.
4. Export owner, environment and last-modified date for each hit.

Two limits matter. The query action accepts only delegated user sign-in; service principals and managed identities get a 403. And the Owner column shows the flow's creator, which does not update if ownership changes. Connector inventory is also not available in GCC, GCC High or DoD clouds. For the agent side of the same tenant, see [how to inventory Copilot Studio agents across environments](/blog/inventory-copilot-studio-agents-across-environments).

## How do you scan Make?

Make's [Scenarios API](https://developers.make.com/api-documentation/api-reference/scenarios) gives you two useful fields. `usedPackages` lists the apps a scenario uses. `usedModules`, requested with the `cols[]` parameter, lists each module with its package and module name. For the full step configuration, including HTTP module URLs, call `GET /scenarios/{scenarioId}/blueprint`.

1. List scenarios per organisation and team with `cols[]` set to include `usedPackages` and `usedModules`.
2. Flag scenarios that use an AI app or Make's own AI modules.
3. Pull the blueprint for every scenario that uses the HTTP module and search it for model host names.
4. On Enterprise plans, read [audit logs](https://help.make.com/audit-logs.md) for connection created, updated and reauthorised events. Make keeps them for 12 months, so you can see when an AI connection first appeared.

## How do you scan Zapier?

Zapier exposes less per-step detail to admins than the other two. On Enterprise accounts, admins manage [app access settings](https://help.zapier.com/hc/en-us/articles/45852792795917-Configure-app-access-settings-for-your-Enterprise-account) and can review app connections and the Zaps that use each app on the Apps page.

1. On the Apps page, review every AI app with connections and list the Zaps that use it.
2. Check whether your account uses an allow list (the default for Enterprise accounts created on or after 5 October 2026) or the older block list. Under a block list, any new AI app is usable until someone blocks it.
3. Use restricted actions to limit specific triggers and actions, including API requests.
4. Note that API by Zapier is not governed by app access controls, so review it separately.

The settings also apply to Zapier Agents, and agents that use a prohibited app may be affected when you tighten the list.

## What do you do with the connection inventory?

The connections are where the risk sits. Each AI connection stores a credential, and that credential decides whose quota is spent, whose data terms apply and who can be cut off.

Build one table across all three tools with these columns:

- Workflow name, tool, environment or team, owner
- AI step type: built-in, provider module, or HTTP
- Model provider and connection name
- Credential type: company key, personal key, or platform-managed
- Data sent: which trigger or earlier step feeds the prompt
- Last run date

Then match each credential against the keys found in your cloud accounts. A key that appears in a workflow but in no approved project is a personal or leaked key. Our guide to [finding leaked and unused LLM API keys](/blog/find-leaked-and-unused-llm-api-keys) covers that matching step. Workflows that touch regulated data with a personal key go to the top of the review list.

## How Opencontroller does this

Opencontroller auto-discovers models and API keys across AWS, Azure and Google Cloud, so an Azure OpenAI key used by a flow shows up on the cloud side even before you scan the workflow tool. Workflows you find can be recorded in the governed registry as agents you do not host, with owner and version. To get cost attribution per workflow, point AI steps at the Lyzr Gateway's OpenAI-compatible endpoint instead of a provider directly; routing, fallback, caching and cost attribution then apply. See the [capability reference](/capability-reference) or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Do I need to scan HTTP steps if I already block AI apps?**
Yes. Blocking an AI app or connector does not stop a generic HTTP or webhook step from posting to a model API. Search step configurations for model host names.

**Can a service principal read the Power Platform inventory?**
Not through the "Query Power Platform resources" action. It supports delegated user authentication only; service principals and managed identities receive HTTP 403.

**How far back can I see when an AI connection was added in Make?**
On Enterprise plans, Make audit logs record connection created, updated and reauthorised events and keep them for 12 months.

## Key facts

- AI steps in automation tools take three forms: built-in AI actions, model-provider modules, and generic HTTP calls to model APIs.
- Power Platform inventory records the connectors and operations each cloud flow and agent flow uses; its query action accepts only delegated user authentication.
- Make's Scenarios API exposes usedPackages and usedModules; GET /scenarios/{id}/blueprint returns the full module configuration.
- Make audit logs (Enterprise plans) record connection events and are kept for 12 months.
- Zapier Enterprise accounts created on or after 5 October 2026 default to an app allow list; API by Zapier is not governed by app access controls.
- Opencontroller auto-discovers models and API keys across AWS, Azure and Google Cloud; Opencontroller pricing starts from $250K per deployment per year.

## Related

- [How do you find leaked or unused LLM API keys?](https://opencontroller.lyzr.com/blog/find-leaked-and-unused-llm-api-keys.md)
- [How do you inventory every Copilot Studio agent across Power Platform environments?](https://opencontroller.lyzr.com/blog/inventory-copilot-studio-agents-across-environments.md)
- [How do you find AI features switched on inside SaaS apps?](https://opencontroller.lyzr.com/blog/find-ai-features-inside-saas-apps.md)
- [More on Discovery and shadow agents](https://opencontroller.lyzr.com/blog/cluster/discovery-and-shadow-agents)
