---
title: How do you find AI agents that contractors use on your data?
url: https://opencontroller.lyzr.com/blog/find-ai-agents-used-by-contractors
published: 2026-11-17
updated: 2026-11-17
author: Lyzr
---

# How do you find AI agents that contractors use on your data?

**You usually cannot see inside a contractor's laptop, but you can see what their AI tools do to your systems. Look at guest accounts and their OAuth grants, sign-ins from unmanaged devices, repository tokens and installed apps, and data flows to AI services. Then close the gap with contract terms and an onboarding checklist.**

Contractors, agencies and offshore teams often bring their own AI tools: coding agents, meeting bots, document assistants. Those tools may touch your code, files and tickets with no review. Employee shadow AI is covered in [finding shadow AI agents](/blog/how-to-find-shadow-ai-agents). Contractors need a different approach because you do not manage their devices or their software licences.

## Why is contractor AI harder to see?

Three reasons. Contractors sign in as guests or with separate accounts, so they fall outside many employee reports. Their tools run on devices you do not manage, so endpoint inventory misses them. And their AI subscriptions are paid by their employer, so expense reviews miss them too. What remains visible is the point where their tools touch your systems.

## Which signals reveal contractor AI use?

| Signal source | What to look for | Where to find it |
|---|---|---|
| Identity provider guests | External accounts that consented to third-party apps | Guest user lists and enterprise app consent logs in Microsoft Entra ID; OAuth token audit in Google Workspace |
| OAuth grants | AI meeting, note-taking or document apps granted mail, calendar or file scopes by external users | App consent and token reports |
| Device posture | Sign-ins to sensitive apps from unmanaged or non-compliant devices | Conditional access and sign-in logs |
| Code hosting | Outside collaborators, personal access tokens, installed apps with repo access | Organisation audit log and app installation list |
| Collaboration tools | Bots added to channels or meetings by guests | Workspace app and bot lists; see [AI bots in Slack and Teams](/blog/find-ai-bots-in-slack-and-teams) |
| Network and DLP | Uploads of company files to AI services from company-provided virtual desktops | Proxy and DLP logs |

Start with identity. [Identity provider logs](/blog/discover-ai-agents-through-identity-provider-logs) show which apps external users consented to, and that list usually contains the AI tools.

## How do you run the review, step by step?

1. **Export all external identities.** Include guests, vendor accounts and service accounts created for contractors. Tag each with the contracting company and sponsor.
2. **Export their app consents and tokens.** Filter for apps whose publisher is an AI vendor or whose name suggests transcription, notes, chat or code.
3. **Check device compliance.** List sign-ins by external users from unmanaged devices to code, document and ticket systems.
4. **Review repository access.** List outside collaborators, tokens they created, and apps installed on repositories they can reach. Look for apps that request read access to all code.
5. **Ask.** Send each supplier a short questionnaire: which AI tools touch client data, where data is processed, and whether it is used for training.
6. **Decide per tool.** Approve, replace with a company-provided tool, or block. Record the decision against the supplier.

## What should contracts and onboarding require?

Contracts set the rules that technical controls cannot enforce on someone else's device.

- **Disclosure.** The supplier lists AI tools used on client data and updates the list when it changes.
- **Approved tools only.** Use limited to an approved list, or to tools the client provides.
- **No training on client data.** Explicit prohibition, with the vendor terms to back it up.
- **Data location and retention.** Where prompts and outputs are stored and for how long.
- **Deletion at exit.** Client data removed from AI tools when the engagement ends, with confirmation.
- **Audit right.** The client can ask for evidence of the above.

At onboarding, give contractors a managed environment (a company laptop or virtual desktop) for sensitive work, with the approved AI tools already installed. It is easier to offer a governed tool than to police an ungoverned one. For suppliers that build agents for you, see [governing agents built by vendors](/blog/govern-ai-agents-built-by-vendors).

## How Opencontroller does this

Opencontroller auto-discovers agents, models and API keys across AWS, Azure and Google Cloud and connects to Claude Enterprise and OpenAI Enterprise workspaces, so company-provided AI tools sit in one governed registry. Opencontroller Edge extends control to developer devices, coding agents, skills and local MCP servers on devices IT manages via MDM, which suits contractors given company laptops. Unmanaged devices still need the contractual controls above. To discuss your supplier estate, [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Can we see AI tools a contractor runs on their own laptop?**
Not directly. You can see their effects: OAuth grants, tokens, repository access and sign-ins from unmanaged devices.

**Should contractors be allowed to use AI tools at all?**
Usually yes, with approved tools in a managed environment. Blanket bans tend to push use out of sight.

**Who owns the contractor AI review?**
The supplier's business sponsor owns the decision, with security running the signal review and procurement owning the contract terms.

## Key facts

To find AI agents that contractors use on company data, review four signal sources: guest and external accounts in the identity provider and their OAuth grants; sign-ins from unmanaged devices; repository access such as outside collaborators, personal access tokens and installed apps; and data movement to AI services in proxy and DLP logs. Then close the gap with contract clauses requiring disclosure of AI tools, and an onboarding checklist. Opencontroller by Lyzr discovers agents, models and API keys across AWS, Azure and Google Cloud, and Opencontroller Edge extends control to developer devices and coding agents when IT installs it via MDM. Pricing starts from $250K per deployment per year.

## Related

- [How do you discover AI agents and apps through identity provider logs?](https://opencontroller.lyzr.com/blog/discover-ai-agents-through-identity-provider-logs.md)
- [How do you find AI features switched on inside SaaS apps?](https://opencontroller.lyzr.com/blog/find-ai-features-inside-saas-apps.md)
- [How do you govern AI agents built by vendors and system integrators?](https://opencontroller.lyzr.com/blog/govern-ai-agents-built-by-vendors.md)
- [More on Discovery and shadow agents](https://opencontroller.lyzr.com/blog/cluster/discovery-and-shadow-agents)
