# Cloudflare Personal Agent Controls: Analytics, Access Policies and Limits

Site owners can now distinguish supported personal assistants and set individual access rules, with an important limit on what agent identity proves.

By Paras Madan | 2026-11-05


Cloudflare announced new personal-agent analytics and access controls on November 4. Website operators can see which supported assistants visit their sites and choose different rules for each one. The update gives a business more options when customers arrive through software that browses on their behalf.

The [release changelog](https://developers.cloudflare.com/changelog/post/2026-11-04-personal-agent-policies/) also renames the dashboard setting to AI Crawler and Agent access policies. It builds on the existing separation between search crawlers, training crawlers and agents.

## What the new analytics show

In [Cloudflare's announcement](https://blog.cloudflare.com/personal-agents-visibility-control/), Security Analytics gains a Personal Assistants category. A separate widget lists assistants and their request counts. Selecting one filters the surrounding traffic information, including requested paths, response codes and source networks.

That can help a retailer investigate why an assistant reaches product pages but never completes a purchase. An error at checkout could point to a broken flow or an access rule. Request counts alone cannot tell the business whether the visit produced a useful comparison or an order, so the application still needs its own outcome measurements.

Account Abuse Protection, which remains in early access, adds account-level context. Its views can associate assistants with login and sign-up events and show which agents have acted for an account. Fraud teams can target a suspicious account through a WAF rule using its hashed user ID, instead of blocking that assistant for every customer.

## How to set a personal-agent policy

The [current policy documentation](https://developers.cloudflare.com/bots/additional-configurations/block-ai-bots/) puts the controls under Security, then Settings, then Bot traffic. Open AI Crawler and Agent access policies and choose an overall Agents action, with individual settings for supported personal agents.

The available choices are Allow (do not block), Block on pages with ads, and Block. Search and Training remain separate policies. A publisher can therefore make a different decision about an assistant fetching a page for a reader than about a crawler collecting material for model training.

There is a configuration detail worth checking after a change: personal agents inherit the overall Agents policy. Updating that policy updates the individual agents too, after which you can set their individual actions again. An operator maintaining exceptions should review them whenever the broader policy changes.

The controls apply to supported agents and detected traffic. They should not be treated as proof that every automated visitor will be identified correctly.

## Recognizing an agent does not establish permission

Cloudflare says carrying user context and authorization through an agent is still future work. Recognizing an assistant does not, by itself, establish what a customer asked it to do or which subscription that customer holds.

For an application team, this means the existing authorization checks still need to run at the point of action. Allowing an assistant to reach a website should not grant it permission to change an account, view another customer's records or complete a transaction without the usual checks.

A useful rollout starts with observation. Look at the pages assistants request, identify where they encounter errors, and compare those paths with the tasks customers actually want to complete. Then change a narrow access rule and check the result. That gives the business evidence for a policy without assuming all agent traffic has the same value or risk.

I'm Paras. I work at the intersection of AI, engineering and distribution.

I help technical founders explain infrastructure products and launch updates in clear stories their customers can use.

[Book a 30-minute call](https://cal.com/paras-madan/30min).