---
title: How often should AI agent discovery scans run?
url: https://opencontroller.lyzr.com/blog/how-often-should-ai-agent-discovery-run
published: 2026-11-11
updated: 2026-11-11
author: Lyzr
---

# How often should AI agent discovery scans run?

**Run a full AI agent discovery scan at least daily, and add event-driven detection so new agents, model deployments and API keys are caught within minutes of being created. Scheduled scans give you completeness; cloud audit-log events give you speed. Neither is enough alone.**

Security teams usually ask two things about discovery: how often it runs, and what it can touch. Permissions are covered in [what permissions AI agent discovery needs](/blog/ai-agent-discovery-permissions-aws-azure-google-cloud). This post covers cadence: how to choose a frequency per asset type, how to stay inside API rate limits, and which changes should count as a "delta" worth alerting on.

## Why not just scan every five minutes?

Because full scans are expensive in API calls, not in compute. A full inventory across many accounts, subscriptions and projects means listing resources in every region for every AI service. Cloud providers throttle control-plane APIs, and a discovery tool that hits those limits can slow down other automation sharing the same quotas.

Very frequent full scans also add little. Most AI assets change slowly. An agent definition might change weekly; a cloud account structure changes monthly. What changes fast is creation: someone spins up a new model deployment or agent this afternoon. That is better caught by events than by re-listing everything.

## What is the right cadence for each asset type?

| Asset type | Full scan | Event-driven detection | Why |
|---|---|---|---|
| Cloud accounts, subscriptions, projects | Daily | Organisation-level events | Scope changes rarely, but a missed account hides everything in it |
| Managed agents (e.g. Bedrock agents, Foundry agents, Vertex AI agents) | Daily | Create/update API events | New agents are the main shadow-AI risk |
| Model deployments and endpoints | Daily | Create/update events | Cost and data-residency exposure starts at deployment |
| API keys and credentials for model providers | Daily | Key-creation events | A new key is often the first sign of an unregistered agent |
| IAM roles and permissions attached to agents | Daily | Policy-change events | Feeds posture checks on risky permissions |
| SaaS agent platforms and enterprise AI workspaces | Daily | Platform webhooks where offered | Many platforms expose no cloud audit trail you control |
| Agent configuration (prompts, tools, MCP servers) | Every few hours to daily | Version-change events | Behaviour changes without a new resource |

Treat these as starting points. Regulated environments often move the full scan to every few hours for production accounts and keep daily for sandboxes.

## How does event-driven detection work in each cloud?

Each cloud writes control-plane activity to an audit log you can subscribe to:

- **AWS:** [CloudTrail](https://docs.aws.amazon.com/awscloudtrail/latest/userguide/cloudtrail-user-guide.html) records management events such as creating an agent or a model endpoint. Routing those events through EventBridge lets you react in near real time instead of waiting for log-file delivery.
- **Azure:** the [Activity log](https://learn.microsoft.com/en-us/azure/azure-monitor/platform/activity-log) records control-plane writes for each subscription and can be exported to Event Hubs or Log Analytics.
- **Google Cloud:** [Cloud Audit Logs](https://cloud.google.com/logging/docs/audit) record Admin Activity, which is always on, and can be routed with a log sink to Pub/Sub.

The pattern is the same in all three: subscribe to create, update and delete events for AI services and IAM, then trigger a targeted re-scan of just that resource. That keeps API usage low and latency short.

## How do you stay inside API rate limits?

1. **Scan incrementally.** After the first full inventory, re-list only accounts and regions where events show change, plus a rolling slice of everything else.
2. **Spread the schedule.** Stagger accounts across the scan window instead of starting them all at midnight.
3. **Back off on throttling.** Honour throttling responses with exponential backoff and jitter; never retry in a tight loop.
4. **Skip empty regions.** Record which regions have never contained AI resources and check them less often, but not never.
5. **Use aggregated inventory APIs** where they exist (for example, organisation-wide asset inventory services) instead of per-account listing.
6. **Use a dedicated identity** for discovery so its API usage is visible and does not compete invisibly with deployment pipelines.

## Which changes should count as a delta?

Not every difference between two scans deserves an alert. Useful delta rules:

- **New asset with no registered owner** → alert and open a review. See [what to do with a shadow AI agent](/blog/what-to-do-with-a-shadow-ai-agent-once-you-find-it).
- **New model provider API key** → alert if it is not linked to a registered agent.
- **Permission broadened** on an agent role (new write or admin actions) → posture finding.
- **Agent version changed** without a matching registry update → flag for review.
- **Asset disappeared** → mark retired, keep history; do not delete the record.
- **Tag-only or description-only changes** → log, do not alert.

Track two numbers to tune cadence: median time from creation to detection, and the share of assets first found by the scheduled scan rather than by events. If the scheduled scan keeps finding things first, your event coverage has gaps.

## How Opencontroller does this

Opencontroller auto-discovers agents, models and API keys across AWS, Azure and Google Cloud, and connects to third-party agent platforms (Salesforce, SAP, Databricks, HubSpot) and Claude Enterprise and OpenAI Enterprise workspaces. Discovered assets land in a governed registry of agents, versions, tools and MCP servers, and Posture flags risky permissions and configurations. Registration is not runtime takeover: discovery tells you what exists, and how deeply you then observe or enforce depends on the connector and traffic path. For the full discovery approach see [how to find shadow AI agents](/blog/how-to-find-shadow-ai-agents), or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**How often should AI agent discovery run?**
A full scan at least daily, plus event-driven detection from cloud audit logs so new agents, deployments and keys appear within minutes.

**Does frequent discovery scanning hit cloud API rate limits?**
It can. Use incremental scans triggered by audit-log events, stagger accounts, and back off on throttling responses.

**What should trigger an alert after a discovery scan?**
New assets with no owner, new model API keys not tied to a registered agent, broadened permissions and unregistered version changes. Tag-only changes should be logged, not alerted.

## Key facts

Recommended AI agent discovery cadence: a full scan at least daily plus event-driven detection from AWS CloudTrail, Azure Activity log and Google Cloud Audit Logs. Opencontroller by Lyzr auto-discovers agents, models and API keys across AWS, Azure and Google Cloud and connects to Salesforce, SAP, Databricks, HubSpot, Claude Enterprise and OpenAI Enterprise. Opencontroller pricing starts from $250K per deployment per year.

## Related

- [How to find every AI agent in your enterprise (including shadow agents)](https://opencontroller.lyzr.com/blog/how-to-find-shadow-ai-agents.md)
- [What permissions does AI agent discovery need in AWS, Azure and Google Cloud?](https://opencontroller.lyzr.com/blog/ai-agent-discovery-permissions-aws-azure-google-cloud.md)
- [What should you do with a shadow AI agent once you find it?](https://opencontroller.lyzr.com/blog/what-to-do-with-a-shadow-ai-agent-once-you-find-it.md)
- [More on Discovery and shadow agents](https://opencontroller.lyzr.com/blog/cluster/discovery-and-shadow-agents)
